Email alert
2016 Vol. 38, No. 1
Display Method:
2016, 38(1): 1-7.
doi: 10.11999/JEIT150600
Abstract:
On the issues about the robustness in visual object tracking, based on Principal Component Analysis (PCA) and Convolutional Neural Network (CNN), a novel visual tracking algorithm with deep feature, which is acquired from a easily initialized CNN structure, is proposed. First, the original image is processed by affine transformation. Next, layered PCA learning is used to process the normalized size image, the eigenvectors learned by PCA are used to be the filters of a CNN structure to realize initialization. Then, the deep expression of the object is extracted by this CNN structure. Finally, combining particle filter algorithm, a simple logistic regression classifier is used to realize target tracking. The result shows that the deep feature acquired from the easily initialized CNN structure has a better expressivity, it can distinguish the object and background effectively. The proposed algorithm has a better inflexibility to illumination, occlusion, rotation and camera shake, and it exhibits a good robustness and accuracy in many video sequences.
On the issues about the robustness in visual object tracking, based on Principal Component Analysis (PCA) and Convolutional Neural Network (CNN), a novel visual tracking algorithm with deep feature, which is acquired from a easily initialized CNN structure, is proposed. First, the original image is processed by affine transformation. Next, layered PCA learning is used to process the normalized size image, the eigenvectors learned by PCA are used to be the filters of a CNN structure to realize initialization. Then, the deep expression of the object is extracted by this CNN structure. Finally, combining particle filter algorithm, a simple logistic regression classifier is used to realize target tracking. The result shows that the deep feature acquired from the easily initialized CNN structure has a better expressivity, it can distinguish the object and background effectively. The proposed algorithm has a better inflexibility to illumination, occlusion, rotation and camera shake, and it exhibits a good robustness and accuracy in many video sequences.
2016, 38(1): 8-14.
doi: 10.11999/JEIT150479
Abstract:
The registration and fusion are the two essential steps to get a composed image from the multispectral infrared images in the night vision. However, at present these two processes are considered as two independent steps, where the registration error may significantly affect the fusion quality. In this paper, a novel iteration optimization method is proposed to obtain the optimal registration parameter for the following fusion process. Definition index of the region of interest in the fused image is used to improve the register process, and simulated annealing method is used to solve the joint optimization problem. The experimental results show that the proposed method provides a robust stability and performance over several other state-of-the-art methods in the registration accuracy and fusion quality.
The registration and fusion are the two essential steps to get a composed image from the multispectral infrared images in the night vision. However, at present these two processes are considered as two independent steps, where the registration error may significantly affect the fusion quality. In this paper, a novel iteration optimization method is proposed to obtain the optimal registration parameter for the following fusion process. Definition index of the region of interest in the fused image is used to improve the register process, and simulated annealing method is used to solve the joint optimization problem. The experimental results show that the proposed method provides a robust stability and performance over several other state-of-the-art methods in the registration accuracy and fusion quality.
2016, 38(1): 15-22.
doi: 10.11999/JEIT150515
Abstract:
A fast visibility restoration method of single image by progressive scene transmission estimation is proposed. Based on the boundary conditions of scene radiance and the atmosphere scattering model, the initial value of the scene transmission is estimated and refined with recursive bilateral filtering. The restoration result is decomposed into a base layer and a detail layer by gradient domain recursive bilateral filtering. The high visibility?restoration result is obtained with tone mapping of the base layer and detail enhancement. Utilizing Gaussian KD tree, the high dimensional feature space of the original image is subdivided adaptively to obtain the Gaussian sampling and accelerate the restoration computing. Compared with related work, the proposed method preserves edges and details, avoiding the halo effectively. The objective indicators are applied to evaluate the restoration results. The experiment results show that the proposed method is effective and feasible, and the restoration results accord with the real scene better.
A fast visibility restoration method of single image by progressive scene transmission estimation is proposed. Based on the boundary conditions of scene radiance and the atmosphere scattering model, the initial value of the scene transmission is estimated and refined with recursive bilateral filtering. The restoration result is decomposed into a base layer and a detail layer by gradient domain recursive bilateral filtering. The high visibility?restoration result is obtained with tone mapping of the base layer and detail enhancement. Utilizing Gaussian KD tree, the high dimensional feature space of the original image is subdivided adaptively to obtain the Gaussian sampling and accelerate the restoration computing. Compared with related work, the proposed method preserves edges and details, avoiding the halo effectively. The objective indicators are applied to evaluate the restoration results. The experiment results show that the proposed method is effective and feasible, and the restoration results accord with the real scene better.
2016, 38(1): 23-28.
doi: 10.11999/JEIT150546
Abstract:
A rotation-invariant feature descripts method called Rotation Invariant Histogram of Oriented Gradients (RI-HOG) is proposed for automatic target recognition. RI-HOG calculates gradient of image first, then the image window is divided into a set of un-overlapped annular regions, called sells, and the Histogram of Gradient (HoG) is used to calculate a feature vector for each cells. After that the HoG of each circle is accumulated to get the main angle of the target area, and then it is rotated due to the main angle to make a normalization of the main angle. At last, the HoG of each circle after rotating is linked to generate the rotation-invariant target feature vector. Experiment results show that target detection method using RI-HOG can find the target under arbitrary rotations. RI-HOG is a rotation-invariant target feature descriptor.
A rotation-invariant feature descripts method called Rotation Invariant Histogram of Oriented Gradients (RI-HOG) is proposed for automatic target recognition. RI-HOG calculates gradient of image first, then the image window is divided into a set of un-overlapped annular regions, called sells, and the Histogram of Gradient (HoG) is used to calculate a feature vector for each cells. After that the HoG of each circle is accumulated to get the main angle of the target area, and then it is rotated due to the main angle to make a normalization of the main angle. At last, the HoG of each circle after rotating is linked to generate the rotation-invariant target feature vector. Experiment results show that target detection method using RI-HOG can find the target under arbitrary rotations. RI-HOG is a rotation-invariant target feature descriptor.
2016, 38(1): 29-37.
doi: 10.11999/JEIT150557
Abstract:
Existing unsupervised feature learning algorithms usually extract features in RGB color space, but YUV color space is widely adopted in image and video compression standards. In order to take advantage of human visual characteristics and avoid the calculation consumption caused by color space conversion, an unsupervised feature learning approach in YUV space based on sparse autoencoders is presented. First, image patches in YUV space are randomly sampled and whitened, and then are fed into sparse autoencoders to learn local features in an unsupervised way. Considering the characteristic that the luminance channel and chrominance channels are independent in YUV space, a whitening method which treats the luminance and chrominance separately is proposed in the pre-processing step. Finally, features learned over local image patches are convolved with large-size images in order to get global feature activations. Global features are then sent into image classification systems for performance testing. Experimental results reveal that unsupervised feature learning in YUV space achieves similar or even slightly better performance in color image classification compared with that in RGB space as long as the luminance component is whitened properly.
Existing unsupervised feature learning algorithms usually extract features in RGB color space, but YUV color space is widely adopted in image and video compression standards. In order to take advantage of human visual characteristics and avoid the calculation consumption caused by color space conversion, an unsupervised feature learning approach in YUV space based on sparse autoencoders is presented. First, image patches in YUV space are randomly sampled and whitened, and then are fed into sparse autoencoders to learn local features in an unsupervised way. Considering the characteristic that the luminance channel and chrominance channels are independent in YUV space, a whitening method which treats the luminance and chrominance separately is proposed in the pre-processing step. Finally, features learned over local image patches are convolved with large-size images in order to get global feature activations. Global features are then sent into image classification systems for performance testing. Experimental results reveal that unsupervised feature learning in YUV space achieves similar or even slightly better performance in color image classification compared with that in RGB space as long as the luminance component is whitened properly.
2016, 38(1): 38-46.
doi: 10.11999/JEIT150337
Abstract:
The most popular approach in object classification is based on the bag of visual-words model. However, there are several fundamental problems that restricts the performance of this method, such as low time efficiency, the synonym and polysemy of visual words, and the lack of spatial information between visual words. In view of this, an object classification method based on weakly supervised Exact Euclidean Locality Sensitive Hashing (E2LSH) and saliency map weighting is proposed. Firstly, E2LSH is employed to generate a group of visual dictionary by clustering SIFT features of the training dataset, and the selecting process of hash functions is effectively supervised inspired by the random forest ideas to reduce the randomcity of E2LSH. Secondly, Graph-Based Visual Saliency (GBVS) algorithm is applied to detect the saliency map of different images and visual words are weighted according to the saliency prior. Finally, saliency map weighted visual language model is carried out to accomplish object classification. Experimental results on datasets of Caltech-256 and Pascal 2007 indicate that the distinguishability of objects is effectively improved and the proposed method is superior to the state- of-the-art object classification methods.
The most popular approach in object classification is based on the bag of visual-words model. However, there are several fundamental problems that restricts the performance of this method, such as low time efficiency, the synonym and polysemy of visual words, and the lack of spatial information between visual words. In view of this, an object classification method based on weakly supervised Exact Euclidean Locality Sensitive Hashing (E2LSH) and saliency map weighting is proposed. Firstly, E2LSH is employed to generate a group of visual dictionary by clustering SIFT features of the training dataset, and the selecting process of hash functions is effectively supervised inspired by the random forest ideas to reduce the randomcity of E2LSH. Secondly, Graph-Based Visual Saliency (GBVS) algorithm is applied to detect the saliency map of different images and visual words are weighted according to the saliency prior. Finally, saliency map weighted visual language model is carried out to accomplish object classification. Experimental results on datasets of Caltech-256 and Pascal 2007 indicate that the distinguishability of objects is effectively improved and the proposed method is superior to the state- of-the-art object classification methods.
2016, 38(1): 47-52.
doi: 10.11999/JEIT150501
Abstract:
In the practical application, non-rigid point set registration should be robust for noise, occlusion or outliers. In this paper, Gaussian Mixture Model (GMM) and neighborhood structure information are used for the non-rigid point set registration. Gaussian Mixture Model is used to represent the model set, and the transformation is built by using Gaussian radial basis function. The proportion of each Gaussian component is decided by the neighborhood structure information of points. In E-step of the EM algorithm the correspondence is solved, and in M-step the outlier ratio and the closed-form solution of the transformation are calculated. Until convergence the optimal solution is obtained. As compared to the state-of-the-art algorithms, the experiments with synthetic data and real data of the retina images show that the proposed method can improve the robustness and the accuracy.
In the practical application, non-rigid point set registration should be robust for noise, occlusion or outliers. In this paper, Gaussian Mixture Model (GMM) and neighborhood structure information are used for the non-rigid point set registration. Gaussian Mixture Model is used to represent the model set, and the transformation is built by using Gaussian radial basis function. The proportion of each Gaussian component is decided by the neighborhood structure information of points. In E-step of the EM algorithm the correspondence is solved, and in M-step the outlier ratio and the closed-form solution of the transformation are calculated. Until convergence the optimal solution is obtained. As compared to the state-of-the-art algorithms, the experiments with synthetic data and real data of the retina images show that the proposed method can improve the robustness and the accuracy.
2016, 38(1): 53-62.
doi: 10.11999/JEIT150157
Abstract:
To address the online time series prediction problem in CPS (Cyber-Physical System) system, both KAF (Kernel Adaptive Filter) with low computation complexity and adaptive characteristic and FPGA computing system are employed. A novel FPGA implementation of vector processor targeting KAF algorithm is proposed. The parallelized datapath and multi-stage pipeline are introduced to enhance the performance and reduce the power consumption and latency. The microcoding technology is further employed to improve the reusability and extensibility. The classical KAF algorithms are implemented based on the vector processor. Experiments results show that the proposed vector processor improves the execution speed by factors of 22, the power consumption decrease to 1/139, while the latency decrease to 1/26 compared with a CPU, on the condition that the precision meets the requirement.
To address the online time series prediction problem in CPS (Cyber-Physical System) system, both KAF (Kernel Adaptive Filter) with low computation complexity and adaptive characteristic and FPGA computing system are employed. A novel FPGA implementation of vector processor targeting KAF algorithm is proposed. The parallelized datapath and multi-stage pipeline are introduced to enhance the performance and reduce the power consumption and latency. The microcoding technology is further employed to improve the reusability and extensibility. The classical KAF algorithms are implemented based on the vector processor. Experiments results show that the proposed vector processor improves the execution speed by factors of 22, the power consumption decrease to 1/139, while the latency decrease to 1/26 compared with a CPU, on the condition that the precision meets the requirement.
2016, 38(1): 63-70.
doi: 10.11999/JEIT150421
Abstract:
Vision-based agricultural vehicle navigation has become a popular research area of automated guidance, however, crop row detection in high weeds field is still a challenging topic. An image segmentation method mainly based on frequency and color information is proposed to remove weeds. The algorithm is based on total frequency parameters, more total crop frequency, alternation regular of crop rows, Otsu method and color model transformation. The total frequency parameters are obtained from wavelet multi-resolution decomposition. The least square method is used in fitting straight line to detect the crop rows. Experiments show that the algorithm can effectively overcome the high weeds. The average processing time of a single pixels image is 132 ms.
Vision-based agricultural vehicle navigation has become a popular research area of automated guidance, however, crop row detection in high weeds field is still a challenging topic. An image segmentation method mainly based on frequency and color information is proposed to remove weeds. The algorithm is based on total frequency parameters, more total crop frequency, alternation regular of crop rows, Otsu method and color model transformation. The total frequency parameters are obtained from wavelet multi-resolution decomposition. The least square method is used in fitting straight line to detect the crop rows. Experiments show that the algorithm can effectively overcome the high weeds. The average processing time of a single pixels image is 132 ms.
2016, 38(1): 71-79.
doi: 10.11999/JEIT150394
Abstract:
Vertical parameters estimation of forest is one of the most important applications in the Polarimetric Interferometric SAR (PolInSAR). On the basis of the Radom Volume over Ground (RVoG) model, the topographical phase and the tree height in forest area can be successfully estimated with the PolInSAR data. In this paper, the topographical phase and forest heights are obtained with single-baseline Compact Polarimetric Interferometric SAR (C-PolInSAR) data. This paper introduces the complex coherence coefficients and the coherence region of the single-baseline C-PolInSAR. Firstly, line-fitting technique is performed on the basis of the coherence region, then the criterion for estimating the topography, and the only volume de-correlation are proposed with the compact polarimetric SAR interferometry datasets. Finally, the topography and the forest heights are proposed. The simulated L, and P band data is used to validate the proposed method, and the topography and tree heights are obtained successfully. Since the transmit wave polarization state of the compact polarimetric system is not unique, this paper analyzes the influence of the elliptic-polarization with different elliptical parameters on the topographical phase and the forest heights. Results show that elliptic-polarization has a little impact on vertical parameters, which also validates the stability of the estimation method.
Vertical parameters estimation of forest is one of the most important applications in the Polarimetric Interferometric SAR (PolInSAR). On the basis of the Radom Volume over Ground (RVoG) model, the topographical phase and the tree height in forest area can be successfully estimated with the PolInSAR data. In this paper, the topographical phase and forest heights are obtained with single-baseline Compact Polarimetric Interferometric SAR (C-PolInSAR) data. This paper introduces the complex coherence coefficients and the coherence region of the single-baseline C-PolInSAR. Firstly, line-fitting technique is performed on the basis of the coherence region, then the criterion for estimating the topography, and the only volume de-correlation are proposed with the compact polarimetric SAR interferometry datasets. Finally, the topography and the forest heights are proposed. The simulated L, and P band data is used to validate the proposed method, and the topography and tree heights are obtained successfully. Since the transmit wave polarization state of the compact polarimetric system is not unique, this paper analyzes the influence of the elliptic-polarization with different elliptical parameters on the topographical phase and the forest heights. Results show that elliptic-polarization has a little impact on vertical parameters, which also validates the stability of the estimation method.
2016, 38(1): 80-89.
doi: 10.11999/JEIT150402
Abstract:
To solve the problem of two dimensional angles estimation for MIMO radar with cross array, a new reduced-dimensional Direction Of Arrival (DOA) estimation method based on ESPRIT algorithm is proposed. Through the reduced-dimensional matrix design and reduced-dimensional transformation, the high dimensional received data can be transformed into a low dimensional signal space, and the corresponding data redundancy can be removed at the greatest degree. The signal space in a real-value field can be estimated through the unitary transformation of matrix, and parameters can be jointly estimated using ESPRIT algorithm with automatic pairing. The proposed algorithm, obtaining signal to noise ratio gain and snapshots gain, can reduce effectively the dimension of received data and the computational complexity of parameters estimation without costing the aperture of array. Lastly, simulation results verify the correctness of theoretical analysis and the effectiveness of the proposed algorithm.
To solve the problem of two dimensional angles estimation for MIMO radar with cross array, a new reduced-dimensional Direction Of Arrival (DOA) estimation method based on ESPRIT algorithm is proposed. Through the reduced-dimensional matrix design and reduced-dimensional transformation, the high dimensional received data can be transformed into a low dimensional signal space, and the corresponding data redundancy can be removed at the greatest degree. The signal space in a real-value field can be estimated through the unitary transformation of matrix, and parameters can be jointly estimated using ESPRIT algorithm with automatic pairing. The proposed algorithm, obtaining signal to noise ratio gain and snapshots gain, can reduce effectively the dimension of received data and the computational complexity of parameters estimation without costing the aperture of array. Lastly, simulation results verify the correctness of theoretical analysis and the effectiveness of the proposed algorithm.
2016, 38(1): 90-96.
doi: 10.11999/JEIT150505
Abstract:
The micro-Doppler signatures can be utilized to the estimation of the motion and structure parameters of the targets. In this paper, based on the effective point scatterer model of the cone-skirt shaped target, the formulas of micro-Doppler induced by the precession are derived. Since the micro-Doppler curves induced by the precession are in the forms of the multi-stage superimposed sine series for the cone-skirt shaped target, an approach to extract the micro-Doppler frequency based on the parameterized time-frequency analysis is proposed. In this method, the precession frequency is first estimated via the Coherent Single Range Doppler Interferometry (CSRDI) algorithm, then the micro-Doppler curve of each scatterer is estimated based on the parameterized time-frequency analysis, and finally the scatterers echoes can be separated with the band-stop filter. In the simulation experiments, the proposed method is evaluated based on the electromagnetic computation data.
The micro-Doppler signatures can be utilized to the estimation of the motion and structure parameters of the targets. In this paper, based on the effective point scatterer model of the cone-skirt shaped target, the formulas of micro-Doppler induced by the precession are derived. Since the micro-Doppler curves induced by the precession are in the forms of the multi-stage superimposed sine series for the cone-skirt shaped target, an approach to extract the micro-Doppler frequency based on the parameterized time-frequency analysis is proposed. In this method, the precession frequency is first estimated via the Coherent Single Range Doppler Interferometry (CSRDI) algorithm, then the micro-Doppler curve of each scatterer is estimated based on the parameterized time-frequency analysis, and finally the scatterers echoes can be separated with the band-stop filter. In the simulation experiments, the proposed method is evaluated based on the electromagnetic computation data.
2016, 38(1): 97-103.
doi: 10.11999/JEIT150366
Abstract:
Feature extraction is a key step in SAR image target recognition. The existence of speckle and discontinuity makes the conventional methods for natural images difficult to apply. Although Deep Belief Networks (DBNs) can be used to learn feature representations automatically, they work essentially in an unsupervised way, and hence the learned features are task-irrelevant. A new Boltzmann machine called Similarity constrained Restricted Boltzmann Machines (SRBMs) is proposed, which injects the supervised information into learning process through constraint on the similarity of feature vectors. Furthermore, a deep architecture named Similarity constrained DBNs (SDBNs) is constructed by layer-wise stacking of SRBMs. Experimental results show the proposed SDBN is superior to DBN and PCA in SAR image target recognition.
Feature extraction is a key step in SAR image target recognition. The existence of speckle and discontinuity makes the conventional methods for natural images difficult to apply. Although Deep Belief Networks (DBNs) can be used to learn feature representations automatically, they work essentially in an unsupervised way, and hence the learned features are task-irrelevant. A new Boltzmann machine called Similarity constrained Restricted Boltzmann Machines (SRBMs) is proposed, which injects the supervised information into learning process through constraint on the similarity of feature vectors. Furthermore, a deep architecture named Similarity constrained DBNs (SDBNs) is constructed by layer-wise stacking of SRBMs. Experimental results show the proposed SDBN is superior to DBN and PCA in SAR image target recognition.
2016, 38(1): 104-110.
doi: 10.11999/JEIT150381
Abstract:
The traditional highly squint SAR imaging algorithm reduces the difficulty in range migration correction by using the time-domain linear walking, and improves the azimuth focusing effect by applying non-linear scaling algorithm. However, the variable standard factor introduced also causes some difficulty for processing. To tackle this problem, this article proposes a new algorithm of azimuth focusing on basis of DFT filter bank theory. Compared with the traditional non-linear scaling algorithm, the new algorithm can compensate Doppler frequency rate better by no introduction of phase operation. The stability is improved and the calculation needed is also less than the traditional algorithm. The simulation results presented proves the effectiveness of the proposed algorithm.
The traditional highly squint SAR imaging algorithm reduces the difficulty in range migration correction by using the time-domain linear walking, and improves the azimuth focusing effect by applying non-linear scaling algorithm. However, the variable standard factor introduced also causes some difficulty for processing. To tackle this problem, this article proposes a new algorithm of azimuth focusing on basis of DFT filter bank theory. Compared with the traditional non-linear scaling algorithm, the new algorithm can compensate Doppler frequency rate better by no introduction of phase operation. The stability is improved and the calculation needed is also less than the traditional algorithm. The simulation results presented proves the effectiveness of the proposed algorithm.
2016, 38(1): 111-118.
doi: 10.11999/JEIT150444
Abstract:
To enhance the mapping ability of artificial neural networks, by studying the mapping mechanism of hidden layer neurons, a new idea of designing neural networks model based on rotation of qubits in the Bloch sphere is proposed in this paper. In the proposed approach, the samples are linearly transformed to quantum bit phase, and the qubits are rotated about three coordinate axes, respectively. The network parameters of hidden layer are the rotation angles. The spherical coordinates of qubits can be obtained by the projection measurement. The output of hidden layer neurons can be concluded by submitting these coordinates to excitation functions in hidden layer. The general neurons are applied to the output layer. The learning algorithms of the proposed model are designed based on the Levenberg-Marquardt (L-M) algorithm. The experimental results show that the proposed model is superior to the classical (L-M) algorthm in approximation ability, generalization ability, and robust performance.
To enhance the mapping ability of artificial neural networks, by studying the mapping mechanism of hidden layer neurons, a new idea of designing neural networks model based on rotation of qubits in the Bloch sphere is proposed in this paper. In the proposed approach, the samples are linearly transformed to quantum bit phase, and the qubits are rotated about three coordinate axes, respectively. The network parameters of hidden layer are the rotation angles. The spherical coordinates of qubits can be obtained by the projection measurement. The output of hidden layer neurons can be concluded by submitting these coordinates to excitation functions in hidden layer. The general neurons are applied to the output layer. The learning algorithms of the proposed model are designed based on the Levenberg-Marquardt (L-M) algorithm. The experimental results show that the proposed model is superior to the classical (L-M) algorthm in approximation ability, generalization ability, and robust performance.
2016, 38(1): 119-126.
doi: 10.11999/JEIT150476
Abstract:
In a passive Ultra-High Frequency (UHF) Radio Frequency IDentification (RFID) system, when multiple tags choose a same time slot to send information to a reader, tag collision will occur. Generally, the collision is resolved only on a Medium Access Control (MAC) layer. In fact, the collision could be separated on a physical layer, and the efficiency of system identification could be advanced. In the physical layer separation, channel estimation is one of key techniques because good estimation could help to correctly recover the collided signals. Conventional channel estimates work well under the environment of two collided tags. When the number of collided tags is beyond two, however, the conventional channel estimates have more estimation errors. In this paper, a novel channel estimate method is proposed for the passive UHF RFID signal separation on physical layer. The proposed method uses the information of preambles which is a-priori known for a reader and applies a Least-Square (LS) criterion to estimate the channel parameters. From numerical results, the estimation errors of the proposed method are lower than the conventional methods under the number of collided tags is more than two. And, the separation efficiency of the proposed methods is also higher.
In a passive Ultra-High Frequency (UHF) Radio Frequency IDentification (RFID) system, when multiple tags choose a same time slot to send information to a reader, tag collision will occur. Generally, the collision is resolved only on a Medium Access Control (MAC) layer. In fact, the collision could be separated on a physical layer, and the efficiency of system identification could be advanced. In the physical layer separation, channel estimation is one of key techniques because good estimation could help to correctly recover the collided signals. Conventional channel estimates work well under the environment of two collided tags. When the number of collided tags is beyond two, however, the conventional channel estimates have more estimation errors. In this paper, a novel channel estimate method is proposed for the passive UHF RFID signal separation on physical layer. The proposed method uses the information of preambles which is a-priori known for a reader and applies a Least-Square (LS) criterion to estimate the channel parameters. From numerical results, the estimation errors of the proposed method are lower than the conventional methods under the number of collided tags is more than two. And, the separation efficiency of the proposed methods is also higher.
Multiple-scale Structural Similarity Image Quality Assessment Based on Internal Generative Mechanism
2016, 38(1): 127-134.
doi: 10.11999/JEIT150616
Abstract:
In order to improve image information uncertainty measurement of the Multiple-scale Structural SIMilarity (MSSIM), a novel algorithm called iMSSIM based on internal generative mechanism is proposed, combining with Human Visual System (HVS). Firstly, internal generative mechanism based on the Piecewise AutoRegressive (PAR) model decomposes distorted image and the original image into two parts, the predicted part of image content using MSSIM algorithm assessment and image information uncertainty Part using PSNR assessment. Then, Mean Square Error is used as weight to combine the two scores to acquire the overall image quality assessmet results. Experiments performed on benchmark IQA databases demonstrate that the proposed algorithm not only has the best performance in different types of distortion, but also is better than the existing algorithms.
In order to improve image information uncertainty measurement of the Multiple-scale Structural SIMilarity (MSSIM), a novel algorithm called iMSSIM based on internal generative mechanism is proposed, combining with Human Visual System (HVS). Firstly, internal generative mechanism based on the Piecewise AutoRegressive (PAR) model decomposes distorted image and the original image into two parts, the predicted part of image content using MSSIM algorithm assessment and image information uncertainty Part using PSNR assessment. Then, Mean Square Error is used as weight to combine the two scores to acquire the overall image quality assessmet results. Experiments performed on benchmark IQA databases demonstrate that the proposed algorithm not only has the best performance in different types of distortion, but also is better than the existing algorithms.
2016, 38(1): 135-140.
doi: 10.11999/JEIT150597
Abstract:
The current Biclique attack is the only key recovery method for the full AES faster than brute-force, but how to get a new Biclique structure or all Biclique structures for AES has not been resolved. This paper designs algorithms to find all Biclique structures for AES-128 and evaluate the computational complexity or data complexity of corresponding Biclique attacks. Using these algorithms, this paper gives that there are 215 kindsi-differentials to generate 555 Biclique structures of AES-128, presentsi-differential trails with the smallest and the second smallest data complexity, and gets Biclique differentials and matching with the smallest computational complexity and the smallest data complexity respectively.
The current Biclique attack is the only key recovery method for the full AES faster than brute-force, but how to get a new Biclique structure or all Biclique structures for AES has not been resolved. This paper designs algorithms to find all Biclique structures for AES-128 and evaluate the computational complexity or data complexity of corresponding Biclique attacks. Using these algorithms, this paper gives that there are 215 kindsi-differentials to generate 555 Biclique structures of AES-128, presentsi-differential trails with the smallest and the second smallest data complexity, and gets Biclique differentials and matching with the smallest computational complexity and the smallest data complexity respectively.
2016, 38(1): 141-145.
doi: 10.11999/JEIT150620
Abstract:
Chaos wireless digital communication is an important development direction of high security wireless communication in the future. Chaos code synchronization is one core technology. According to characteristics of chaos wireless communication, a method of chaos code synchronization based on sliding correlation is put forward. For non-periodic chaos code synchronization, phase delay controller is designed under the condition of three constrain conditions. Taking logistic sequence for example, a dynamic model of chaos code synchronization system is built and the simulation is carried out. The test results show that this method can effectively realize chaos code synchronization between sender and receiver. Its synchronous rate is fast and anti-noise performance is good. It solves the synchronization problem in chaos wireless digital communication.
Chaos wireless digital communication is an important development direction of high security wireless communication in the future. Chaos code synchronization is one core technology. According to characteristics of chaos wireless communication, a method of chaos code synchronization based on sliding correlation is put forward. For non-periodic chaos code synchronization, phase delay controller is designed under the condition of three constrain conditions. Taking logistic sequence for example, a dynamic model of chaos code synchronization system is built and the simulation is carried out. The test results show that this method can effectively realize chaos code synchronization between sender and receiver. Its synchronous rate is fast and anti-noise performance is good. It solves the synchronization problem in chaos wireless digital communication.
2016, 38(1): 146-152.
doi: 10.11999/JEIT150396
Abstract:
For future railway wireless communication networks, it is an effective way to adopt higher frequency spectra with broader bandwidth to enhance the transmission capacity. Nevertheless, massive beamforming techniques are needed to overcome the severe path loss of higher frequency spectra. For railway systems with dual on-vehicle receivers, dual-beam transmissions can be implemented to improve the capacity. The analysis results show that the optimization of dual-beam transmissions depends on the train position. Based on the above, an adaptive beam splitting or integrating communication scheme is proposed. When the train is far away from the base station, to avoid the inter-beam interference, an integrated beam with wider beamwidth is used to cover the two receivers to realize diversity receiving. As the train is approaching the center of the base station, two beams are generated to realize space multiplexing, to improve the transmission capacity and reliability. Numerical simulation results demonstrate that the proposed scheme can adapt to train positions and improve the transmission performance.
For future railway wireless communication networks, it is an effective way to adopt higher frequency spectra with broader bandwidth to enhance the transmission capacity. Nevertheless, massive beamforming techniques are needed to overcome the severe path loss of higher frequency spectra. For railway systems with dual on-vehicle receivers, dual-beam transmissions can be implemented to improve the capacity. The analysis results show that the optimization of dual-beam transmissions depends on the train position. Based on the above, an adaptive beam splitting or integrating communication scheme is proposed. When the train is far away from the base station, to avoid the inter-beam interference, an integrated beam with wider beamwidth is used to cover the two receivers to realize diversity receiving. As the train is approaching the center of the base station, two beams are generated to realize space multiplexing, to improve the transmission capacity and reliability. Numerical simulation results demonstrate that the proposed scheme can adapt to train positions and improve the transmission performance.
2016, 38(1): 153-159.
doi: 10.11999/JEIT150415
Abstract:
ProtoGraph-based Low Density Parity Check (PG-LDPC) codes have many advantages over the conventional LDPC codes, such as simple structure, low iterative decoding threshold, easy extension and linear encoding/decoding complexity. After investigating the characteristics of PG-LDPC codes over the Shallow Water Acoustic (SWA) channels, which have the features of strong multipath interference, long delay spreading and limited bandwidth, a new design scheme to search for good codes is proposed. Furthermore, protograph- degree-distribution-based EXtrinsic Information Transfer (EXIT) chart algorithm is used to predict and analyze the error performances of protograph-based LDPC codes. The simulation and experiment results show that the proposed code outperforms the (3, 6) randomly regular LDPC code in both low and high SNR region, over the SWA channels.
ProtoGraph-based Low Density Parity Check (PG-LDPC) codes have many advantages over the conventional LDPC codes, such as simple structure, low iterative decoding threshold, easy extension and linear encoding/decoding complexity. After investigating the characteristics of PG-LDPC codes over the Shallow Water Acoustic (SWA) channels, which have the features of strong multipath interference, long delay spreading and limited bandwidth, a new design scheme to search for good codes is proposed. Furthermore, protograph- degree-distribution-based EXtrinsic Information Transfer (EXIT) chart algorithm is used to predict and analyze the error performances of protograph-based LDPC codes. The simulation and experiment results show that the proposed code outperforms the (3, 6) randomly regular LDPC code in both low and high SNR region, over the SWA channels.
2016, 38(1): 160-167.
doi: 10.11999/JEIT150510
Abstract:
Traditional chaotic detection methods have many problems, such as low criterion accuracy and delay state response. To cope with these problems, a weak signal detection method based on dominative frequency power ratio derived from systems first-order perturbation solution is proposed in this paper. This algorithm is ascribable to the all-around analyses of chaotic states global property and system solutions frequency-domain characteristics. It not only gives an effective and accurate critical threshold which could offer more reliable guarantee for signal detection, but also disclosures the differences between system states and the coherent physical meanings. The first-order perturbation equilibrium solution of Duffing-Van der pol oscillator is derived with parameter perturbation method, and it is proved that this solutionis is most significant to the dominative frequency. And then, the effective signal is selectively reconstructed through empirical mode decomposition, and system state is redefined with this ratio restrained under MMSE criterion. Finally the mapping relationship between power ratio of dominative frequencies and driving motivation amplitude is obtained and it is considered as determination criterion of critical threshold. Experimental results show that this algorithm could bring an promotion about one order of magnitude in system reliability, and the response speed is at least doubled compared with traditional methods.
Traditional chaotic detection methods have many problems, such as low criterion accuracy and delay state response. To cope with these problems, a weak signal detection method based on dominative frequency power ratio derived from systems first-order perturbation solution is proposed in this paper. This algorithm is ascribable to the all-around analyses of chaotic states global property and system solutions frequency-domain characteristics. It not only gives an effective and accurate critical threshold which could offer more reliable guarantee for signal detection, but also disclosures the differences between system states and the coherent physical meanings. The first-order perturbation equilibrium solution of Duffing-Van der pol oscillator is derived with parameter perturbation method, and it is proved that this solutionis is most significant to the dominative frequency. And then, the effective signal is selectively reconstructed through empirical mode decomposition, and system state is redefined with this ratio restrained under MMSE criterion. Finally the mapping relationship between power ratio of dominative frequencies and driving motivation amplitude is obtained and it is considered as determination criterion of critical threshold. Experimental results show that this algorithm could bring an promotion about one order of magnitude in system reliability, and the response speed is at least doubled compared with traditional methods.
2016, 38(1): 168-173.
doi: 10.11999/JEIT150538
Abstract:
A novel Direction-Of-Arrival (DOA) estimation algorithm based on spatial smoothing and sparse reconstruction is proposed in this paper. Firstly, the covariance matrix is processed using spatial smoothing theory, and it is converted with the Khatri-Rao transformation, then DOA estimation is achieved by sparse reconstruction of the converted matrix. Furthermore, two different kinds of methods are given to deal with the error of the objective function. Experimental results show that the proposed algorithm can reduce the amount of computation, and exhibit better performance on both coherent and non-coherent signals compared with the other DOA algorithms based on compressed sensing, especially under the conditions of low angle interval, low signal-to-noise ratio and low sampling number.
A novel Direction-Of-Arrival (DOA) estimation algorithm based on spatial smoothing and sparse reconstruction is proposed in this paper. Firstly, the covariance matrix is processed using spatial smoothing theory, and it is converted with the Khatri-Rao transformation, then DOA estimation is achieved by sparse reconstruction of the converted matrix. Furthermore, two different kinds of methods are given to deal with the error of the objective function. Experimental results show that the proposed algorithm can reduce the amount of computation, and exhibit better performance on both coherent and non-coherent signals compared with the other DOA algorithms based on compressed sensing, especially under the conditions of low angle interval, low signal-to-noise ratio and low sampling number.
2016, 38(1): 174-179.
doi: 10.11999/JEIT150504
Abstract:
In order to enhance the whole quality of single channel speech enhancement denoising algorithm, both noise reducing and speech perception are considered to improve the traditional speech enhancement algorithm and many kinds of processing methods are taken to achieve the best optimization effect. Firstly, in the view of parameters estimation, spectrum smoothing algorithm based on weak speech presence is added to the soft decision method based on fixed prior signal-to-noise ratio in order to solve the problem of noise spectrum overestimation. Moreover, the smoothing parameter is dynamically controlled by the speech presence probability in order to enhance the tracing effect of prior signal-to-noise ratio. Secondly, in the view of the speech perception improvement, the harmonic reconstruction method is used to reconstruct the harmonic components in high frequencies of speech section. Phase compensation method and gain smoothing method are also employed to remove the annoying musical noise in speech and silence segment. The experimental results show that compared with the traditional algorithm, the proposed algorithm obtains good performance in both denoising effect and speech quality by introducing parameter estimation improvement module and perceived quality improvement module, and it is suitable for many kinds of noise environment and signal-to-noise ratio conditions.
In order to enhance the whole quality of single channel speech enhancement denoising algorithm, both noise reducing and speech perception are considered to improve the traditional speech enhancement algorithm and many kinds of processing methods are taken to achieve the best optimization effect. Firstly, in the view of parameters estimation, spectrum smoothing algorithm based on weak speech presence is added to the soft decision method based on fixed prior signal-to-noise ratio in order to solve the problem of noise spectrum overestimation. Moreover, the smoothing parameter is dynamically controlled by the speech presence probability in order to enhance the tracing effect of prior signal-to-noise ratio. Secondly, in the view of the speech perception improvement, the harmonic reconstruction method is used to reconstruct the harmonic components in high frequencies of speech section. Phase compensation method and gain smoothing method are also employed to remove the annoying musical noise in speech and silence segment. The experimental results show that compared with the traditional algorithm, the proposed algorithm obtains good performance in both denoising effect and speech quality by introducing parameter estimation improvement module and perceived quality improvement module, and it is suitable for many kinds of noise environment and signal-to-noise ratio conditions.
2016, 38(1): 180-186.
doi: 10.11999/JEIT150425
Abstract:
In a conventional narrowband active noise control system, each narrowband component is cancelled separately, but each control coefficient is still updated by the overall residual error of the system. This leads to the interferences among the controllers, thus reducing the convergence rate of the system. In order to ensure that each controller uses its own residual error to update the control coefficient and to separate frequency components from residual error signals, a complete narrowband active noise control system based on leaky residual error separator is proposed, and the corresponding preliminary statistical analysis on the new system is derived. Extensive simulations and experimental results as well as theoretical analysis demonstrate that the robustness and convergence performances can be improved without increasing the residual noise in the steady state by using the residual error separator that introduces the leakage factor.
In a conventional narrowband active noise control system, each narrowband component is cancelled separately, but each control coefficient is still updated by the overall residual error of the system. This leads to the interferences among the controllers, thus reducing the convergence rate of the system. In order to ensure that each controller uses its own residual error to update the control coefficient and to separate frequency components from residual error signals, a complete narrowband active noise control system based on leaky residual error separator is proposed, and the corresponding preliminary statistical analysis on the new system is derived. Extensive simulations and experimental results as well as theoretical analysis demonstrate that the robustness and convergence performances can be improved without increasing the residual noise in the steady state by using the residual error separator that introduces the leakage factor.
2016, 38(1): 187-194.
doi: 10.11999/JEIT150437
Abstract:
In 3D forward problem of Magnetic Induction Tomography (MIT), the problems are slow computation speeds and incorrect results due to round-off errors, when calculating the finite element equations with the direct method. Incomplete Cholesky Conjugate Gradient (ICCG) iteration method is used to solve these problems. Round-off errors are compensated by iteration method. An Finite-Element Model (FEM) is built based on the ANSYS software. The FEM equations are solved by the ICCG method. The optimal convergence tolerance value is calculated. Simulation result shows that the ICCG method has advantages in speed and stability compared with direct and Jacobi Conjugate Gradient (JCG) method. The results show that the ICCG method is not affected by meshing perturbation, it can solve the 3D forward problem of MIT correctly.
In 3D forward problem of Magnetic Induction Tomography (MIT), the problems are slow computation speeds and incorrect results due to round-off errors, when calculating the finite element equations with the direct method. Incomplete Cholesky Conjugate Gradient (ICCG) iteration method is used to solve these problems. Round-off errors are compensated by iteration method. An Finite-Element Model (FEM) is built based on the ANSYS software. The FEM equations are solved by the ICCG method. The optimal convergence tolerance value is calculated. Simulation result shows that the ICCG method has advantages in speed and stability compared with direct and Jacobi Conjugate Gradient (JCG) method. The results show that the ICCG method is not affected by meshing perturbation, it can solve the 3D forward problem of MIT correctly.
2016, 38(1): 195-201.
doi: 10.11999/JEIT150491
Abstract:
In order to solve the problem of connecting the wireless sensor network with the Internet in Cyber- physical systems, a gateway deployment algorithm based on differential evolution is proposed. This algorithm uses the differential evolution algorithm to optimize the minimum coverage radius and gateway load balancing. With the improvement of adaptive opposition-based search and dynamic parameters adjustment, this algorithm can keep the variety of the whole swarm and solve the geometric -center problem. Simulation results show that, this algorithm gets good global explorative ability and convergence speed, and can benefit the network QoS level of the Cyber-physical systems by obtaining good load balancing and minimum coverage radius.
In order to solve the problem of connecting the wireless sensor network with the Internet in Cyber- physical systems, a gateway deployment algorithm based on differential evolution is proposed. This algorithm uses the differential evolution algorithm to optimize the minimum coverage radius and gateway load balancing. With the improvement of adaptive opposition-based search and dynamic parameters adjustment, this algorithm can keep the variety of the whole swarm and solve the geometric -center problem. Simulation results show that, this algorithm gets good global explorative ability and convergence speed, and can benefit the network QoS level of the Cyber-physical systems by obtaining good load balancing and minimum coverage radius.
2016, 38(1): 202-208.
doi: 10.11999/JEIT150247
Abstract:
The large scale and the coarse classification granularity of resources in literature knowledge bases lead to disorientation and overloading when learners retrieve and read papers. This paper proposes a mechanism of knowledge clustering and knowledge statistics based on MapReduce. Firstly, this paper presents a Co-occurrence Matrix building algorithm based on MapReduce (MR-CoMatrix). Secondly, it makes combination of the co-occurrence matrix and similarity coefficient to build the similarity matrix. Thirdly, the similarity matrix is standardized with Z scores. Finally, knowledge clusters are constructed with the Ward,s method. After knowledge clustering, this paper introduces a knowledge Statistics algorithm based on MapReduce (MR-Statistics) to dig the hidden information in each cluster. The experimental results show that the literature knowledge base with MR- CoMatrix and MR-Statistics can realize the accurate and fine clustering, multi-dimension statistics, computational efficiency, and less cost of time.
The large scale and the coarse classification granularity of resources in literature knowledge bases lead to disorientation and overloading when learners retrieve and read papers. This paper proposes a mechanism of knowledge clustering and knowledge statistics based on MapReduce. Firstly, this paper presents a Co-occurrence Matrix building algorithm based on MapReduce (MR-CoMatrix). Secondly, it makes combination of the co-occurrence matrix and similarity coefficient to build the similarity matrix. Thirdly, the similarity matrix is standardized with Z scores. Finally, knowledge clusters are constructed with the Ward,s method. After knowledge clustering, this paper introduces a knowledge Statistics algorithm based on MapReduce (MR-Statistics) to dig the hidden information in each cluster. The experimental results show that the literature knowledge base with MR- CoMatrix and MR-Statistics can realize the accurate and fine clustering, multi-dimension statistics, computational efficiency, and less cost of time.
2016, 38(1): 209-215.
doi: 10.11999/JEIT150210
Abstract:
Study on the relativities of the tropospheric microwave trans-horizon propagation above ocean surface and the marine atmospheric environment characteristics has important significance for the short-term prediction of the trans-horizon propagation and for the designs and the applications of the radio-communication systems. In this paper, based on the transmission loss data collected in the oversea experiment at 14.1 GHz on the area of Yellow Sea and Bohai Sea of China and the synchronous meteorological data collected from the meteorology grads tower which founded in the transmitter station, the relativities of the transmission loss and the evaporation duct height are analyzed with wind direction, wind speed and the difference of air temperature and sea temperature, respectively. The usability of the coastal meteorological data for the tropospheric microwave trans-horizon propagation is studied, and the results are validated with the parabolic equation method and the Advance Refractive Effects Prediction System (AREPS). The conclusions are helpful for the study of the propagation characteristic and the short-term prediction of the tropospheric microwave trans-horizon propagation above ocean surface.
Study on the relativities of the tropospheric microwave trans-horizon propagation above ocean surface and the marine atmospheric environment characteristics has important significance for the short-term prediction of the trans-horizon propagation and for the designs and the applications of the radio-communication systems. In this paper, based on the transmission loss data collected in the oversea experiment at 14.1 GHz on the area of Yellow Sea and Bohai Sea of China and the synchronous meteorological data collected from the meteorology grads tower which founded in the transmitter station, the relativities of the transmission loss and the evaporation duct height are analyzed with wind direction, wind speed and the difference of air temperature and sea temperature, respectively. The usability of the coastal meteorological data for the tropospheric microwave trans-horizon propagation is studied, and the results are validated with the parabolic equation method and the Advance Refractive Effects Prediction System (AREPS). The conclusions are helpful for the study of the propagation characteristic and the short-term prediction of the tropospheric microwave trans-horizon propagation above ocean surface.
2016, 38(1): 216-221.
doi: 10.11999/JEIT150553
Abstract:
Space-Time Adaptive Processing (STAP) is effective to suppress wideband jammers in satellite navigation system. But in high-dynamic environment, the conventional STAP anti-jamming algorithms are invalid since jammers may easily move out of the array pattern null so that it can not be suppressed. In this paper, a new method of STAP null-widen is deduced based on Laplace distribution model of the changing interference DOA in high-dynamic environment. This method can broaden the width of nulls. But because of using of STAP, the amount of computation is increased significantly, a STAP null-widen method based on reduced-dimension Multistage Wiener Filters (MWF) is given in this paper. The effectiveness of the new method is proved in simulation part.
Space-Time Adaptive Processing (STAP) is effective to suppress wideband jammers in satellite navigation system. But in high-dynamic environment, the conventional STAP anti-jamming algorithms are invalid since jammers may easily move out of the array pattern null so that it can not be suppressed. In this paper, a new method of STAP null-widen is deduced based on Laplace distribution model of the changing interference DOA in high-dynamic environment. This method can broaden the width of nulls. But because of using of STAP, the amount of computation is increased significantly, a STAP null-widen method based on reduced-dimension Multistage Wiener Filters (MWF) is given in this paper. The effectiveness of the new method is proved in simulation part.
2016, 38(1): 222-228.
doi: 10.11999/JEIT150481
Abstract:
Continuous Wave Interference (CWI) can induce large code tracking error, and the tracking error analysis based on small error conditions becomes invalid. Under large error conditions, since code loop discriminator is no longer working in the linear region, the second-order Taylor series expansion of discriminator output is introduced instead of linear approximation, and analytical expressions of the tracking error in the presence of CWI are derived. Numerical analysis and simulation results are presented to evaluate the code tracking error with different initial phases，interference frequencies and interference-to-signal power ratio (ISR). The experimental results show that, when the code tracking error induced by CWI is large, the prediction accuracy of theoretical analysis based on linearized discriminator decreases noticeable, while the new theoretical expressions based on second-order Taylor series expansion provide accurate predictions of code tracking errors under large error conditions. If the code tracking error is smaller than 0.34 chip, the theoretical prediction error is no more than 20%.
Continuous Wave Interference (CWI) can induce large code tracking error, and the tracking error analysis based on small error conditions becomes invalid. Under large error conditions, since code loop discriminator is no longer working in the linear region, the second-order Taylor series expansion of discriminator output is introduced instead of linear approximation, and analytical expressions of the tracking error in the presence of CWI are derived. Numerical analysis and simulation results are presented to evaluate the code tracking error with different initial phases，interference frequencies and interference-to-signal power ratio (ISR). The experimental results show that, when the code tracking error induced by CWI is large, the prediction accuracy of theoretical analysis based on linearized discriminator decreases noticeable, while the new theoretical expressions based on second-order Taylor series expansion provide accurate predictions of code tracking errors under large error conditions. If the code tracking error is smaller than 0.34 chip, the theoretical prediction error is no more than 20%.
2016, 38(1): 229-240.
doi: 10.11999/JEIT151116
Abstract:
Throughout the progress of spaceborne Synthetic Aperture Radar (SAR) technology, it is switching from traditional specific technology breakthrough to concept and system innovation. Nowadays various application- oriented new systems and new models are emerging, which push forward the development of spaceborne SAR technology. This paper provides an introduction on the status of spaceborne SAR technology in Europe, the United States and other countries and discusses the trends of future spaceborne SAR technology. Emphases are laid on the development of spaceborne SAR technology for high-resolution wide-swath observation, multi-azimuth information acquisition, high-temporal information acquisition, 3-D terrain mapping and image quality improvement.
Throughout the progress of spaceborne Synthetic Aperture Radar (SAR) technology, it is switching from traditional specific technology breakthrough to concept and system innovation. Nowadays various application- oriented new systems and new models are emerging, which push forward the development of spaceborne SAR technology. This paper provides an introduction on the status of spaceborne SAR technology in Europe, the United States and other countries and discusses the trends of future spaceborne SAR technology. Emphases are laid on the development of spaceborne SAR technology for high-resolution wide-swath observation, multi-azimuth information acquisition, high-temporal information acquisition, 3-D terrain mapping and image quality improvement.
2016, 38(1): 241-245.
doi: 10.11999/JEIT150694
Abstract:
This paper proposes Regularized Semi-Supervised ISOmetric MAPping (Reg-SS-ISOMAP) algorithm to solve the problem that ISOmetric MAPping (ISOMAP) algorithm is unsupervised and can not generate explicit mapping function. At first, this algorithm creates K-Connectivity Graph (K-CG) by labeled samples in training samples to get geodesic distance between approximate samples and takes it as feature vector substituting for original data. Then, it takes the geodesic distance as kernel and processes feature vector through semi-supervised regularization not MultiDimensional Scaling (MDS) algorithm. At last, it constructs objective function by regularization regression model which is low dimension and explicit mapping. The algorithm is simulated on different data sets, results show that it is stable in dimension reduction and high recognition rate.
This paper proposes Regularized Semi-Supervised ISOmetric MAPping (Reg-SS-ISOMAP) algorithm to solve the problem that ISOmetric MAPping (ISOMAP) algorithm is unsupervised and can not generate explicit mapping function. At first, this algorithm creates K-Connectivity Graph (K-CG) by labeled samples in training samples to get geodesic distance between approximate samples and takes it as feature vector substituting for original data. Then, it takes the geodesic distance as kernel and processes feature vector through semi-supervised regularization not MultiDimensional Scaling (MDS) algorithm. At last, it constructs objective function by regularization regression model which is low dimension and explicit mapping. The algorithm is simulated on different data sets, results show that it is stable in dimension reduction and high recognition rate.
