摘要:
以像素值为基础的传统图像质量评价方法有其固有局限性，如对图像结构的忽视及对完全参考图像的需求等。为解决这些问题，该文研究了图像的奇异值向量对图像结构的表征能力，提出了基于视觉权重的奇异值分解和均值偏差率的部分参考图像质量评价方法BWSVD(Block Weighted Singular Value Decomposition)。首先，将图像分成88 大小的图像块，再利用其奇异值向量差值和均值偏差来定量描述图像畸变程度，并结合人眼视觉敏感性为每个图像块赋予一个视觉权重。最后，利用the Live Image Quality Assessment Database, Release2005中的227幅不同压缩倍率的JPEG2000降质图像进行实验，并与PSNR，RMSE，UQI，MSSIM，MSVD等算法进行了对比，实验表明，该文算法对压缩图像质量评价具有更好的稳定性，同时体现了更好的主客观评价一致性。
Abstract:
The traditional image quality assessment methods based on pixels have their own limitations, such as the lack of consideration of the image structure, or the need of a complete reference image. To avoid these problems, this paper presented a new image quality assessment method based on Block Weighted Singular Value Decomposition (BWSVD). First, the images are divided into blocks at size 88, then, the singular value vector difference and the mean bias between the original image blocks and the distorted image blocks are considered to evaluate the distortion degree. Beside this, the Human Visual Sensibility (HVS) is considered to determine the weight of each block. Many tests are conducted to evaluate the performance; the 227 testing images are coming from the Live Image Quality Assessment Database, Release 2005. Compared with the PSNR, RMSE, UQI, MSSIM, MSVD algorithms, the presented method shows a great improvement in both the consistency with the DMOS (Differential Mean Opinion Score, DMOS) and the stability when it is applied to different compression rates.
下载:
