三种基于空间转换的彩色图象分割算法比较研究  

Color Image Segment Methods Based on Color Space Conversion

作　　者：张笃振[1] 刘淑娥[2] 

机构地区：[1]徐州师范大学计算机科学与技术学院,江苏徐州221116 [2]徐州师范大学物理系,江苏徐州221116

出　　处：《电脑知识与技术》2006年第7期146-146,188,共2页Computer Knowledge and Technology

基　　金：徐州师范大学自然科学研究基金项目（04XLB28）.

摘　　要：本文讨论了彩色图象在空闻转换中的分割方法。在RGB空间中，由于R、G、B分量之闻有很高的相关性，直接利用这些分量不能得到所需的效果，探讨了经过正交变换后的分割，实验结果表明分割效果较好。HSI彩色空间和YCbCr彩色空间都具有把亮度分量与其它分量分离的优点。在样本集协方差矩阵的特征向量构成的空间中进行分割效果更好．表明肤色点经转换后在这一肤色空间中具有良好的聚类特性，在YCbCr空间中分割效果较差。在三种分割方法中，如果图象中含有大面积的与肤色十分相似的区域，分割的结果将不尽如人意，这可与人脸的其它特征相结合，最终实现肤色的正确分割。In this article,segment methods of color image in color space conversion are discussed.In RGB color space,because of highly correlated among R、G and B components,direcdy use these components to segment is not effect,but after a orthononnal transform, the experiment result is better.HSI space and YCbCr space both have the advantages of intensity independent of the other two components,experiment result show in HSI space the segmentation is perfect and in YCbCr is not good, this can be summarized that skin-color points after transformed have satisfying cluster property in the skin-color space formed by the eigenvectors of covariance matrix which is trained by a set of points of skin-color.If images have large areas which are similar to skin-color, the result is bad,this can be changed by combining with other face features.

关 键 词：彩色图象分割 RGB空间 HSI空间 YCBCR空间 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]