利用Relief特征加权进行基于小波变换的纹理分割

Wavelet Transform-based Texture Segmentation Using Feature Weighted by Relief

在利用小波进行纹理分割的相关研究中,通常小波分解的四个子带对分类的贡献是均匀的。为了考虑不同子带对图像分割的不同影响,提出了一种利用Relief算法对小波分解的子带特征进行加权的算法。首先对纹理图像进行标准金字塔结构小波变换,对小波变换后的各层四个子带进行特征提取作为纹理图像的四维特征;从粗尺度开始对纹理图像进行K均值分割,得到初步分割结果;然后把初步分割结果扩展到下一尺度,根据扩展后的分割标记图对相应尺度的纹理特征进行基于Relief的特征加权,得到加权后的四维特征;再进行K均值分割,经过多层迭代后,得到原纹理图像的分割结果。实验结果表明,与未加权的传统分割方法比较,该方法在分割错误率、边缘准确性以及区域一致性上都有明显改善。

In the field of texture segmentation method based on wavelet analysis, ordinary, the each layers of the wavelet transform plays a uniform contribution for segmentation. To consider the particular contribution of differ-ent layers, a new approach to weight the features of each layers is presented. Firstly, the standard pyramid-struc-tured wavelet transform is employed. Features from the four bands of the each layers of the wavelet transform forms 4-dimensional features is extracted. The segmentation starts at the lowest resolution using the K-means clustering scheme and the result is propagated to a higher one. According the propagated result , the 4-dimensional features of higher resolution are weighted by relief algorithm , then using the K-means clustering. After multiple iterations, the final segmented image is got. Result shows that compared with a typical traditional method without feature weigh-ting, the present approach shows visible improvements both in diminishing segmentation error, and in increasing boundary precision and region harmony.