改进的同心离散圆簇形状描述方法
doi: 10.3724/SP.J.1146.2012.01526
An Improved Shape Descriptor of Cluster of Concentric Discrete Circles
-
摘要: 形状特征是MPEG7中用来描述图像的重要特征之一。同心离散圆簇(Cluster of Concentric Discrete Circles, CCDC)形状特征提取方法具有特征提取速度快的优点。但该方法的特征函数使用圆环弧段的方差作为特征,由于内外环的离散点数不同,使得内外环的特征值变化范围不同,会产生特征掩盖现象。针对上述问题,该文对特征函数进行了重新设计,利用弧段的相对弧长或段数作为特征,将每一环的特征用4个子特征来描述,并将每一个特征值归一化为0~1范围内,提取过程比原来更简单,特征提取速度更快。改进后的新方法命名为改进的CCDC (Improved CCDC, ICCDC)。该文采用了MPEG7-CE1-B标准形状数据库进行测试,评价指标采用Precision-Recall曲线。实验表明,ICCDC比CCDC在性能方面有显著的提升,检索精度比原来提高了约50%,提取MPEG7-CE1-B图形库中所有图形特征所用的计算时间比原来减少了约25 ms。Abstract: Shape feature is one of the important characteristics which are used to describe images in MPEG7. The descriptor of Cluster of Concentric Discrete Circles (CCDC) can be extracted in a short time. However, the method uses the variance of the ring arc segment as characteristic in its feature function; it is because that the discrete points of inner and outer circle are different. The feature values range of inner and outer circle are also different, and then it leads to produce characteristic covered up phenomenon. For the above problems, the feature functions are redesigned in this paper. The relative arc length of the arc segment or segments is used as a feature. Each ring characterized is expressed by four sub-features, and each feature value is normalized to the range of 0~1, the extraction process is more simple and faster than that of the original CCDC descriptor. The new method is called Improved CCDC (ICCDC). In this paper, all experiments are done by using the MPEG7-CE1-B standard shape database, and the performance is evaluated by the Precision-Recall curve. The experiments show that the performance of the improved CCDC is better than that of the original method. And the retrieval precision is 50% higher than that of the original method, and the time of extracting all the shape in MPEG7-CE1-B is reduced about 25 ms than that of original method.
-
Key words:
- Image processing /
- Shape feature /
- Feature extraction /
- Moment invariants /
- Zernike moments
计量
- 文章访问数: 2710
- HTML全文浏览量: 135
- PDF下载量: 926
- 被引次数: 0