基于小波域多尺度Markov随机场的织物印花图案分割

Textile printing pattern image segmentation based on multiscale Markov random field model in wavelet domain

为提高纺织印花图案的分色精度,采用小波域的多尺度Markov随机场模型实现了织物印花图案的分割。首先对图像进行预处理,然后对图像进行小波金字塔分解,在每个尺度的分割过程中利用了各尺度上的有关信息:特征场建模通过描述小波系数之间的相关性反映每个像素位置的特征属性,标记场建模通过考虑邻域标记间的相互作用反映图像分割的区域性。2种随机场建模以联合概率乘积的形式相互约束,共同作用于该尺度的分割过程。在分割过程中,从最低分辨率尺度到原始分辨率尺度逐次进行图像分割,低分辨率尺度的分割结果通过直接投影作为相邻的更高分辨率尺度的初始分割,最高分辨率尺度上的分割结果作为该方法的分割结果。

To improve the color separation accuracy of textile printing patterns, this paper adopted MRF algorithm and realized texture segmentation. At first a preceding process is taken in this algorithm. Then, the image is decomposed by wavelet pyramids. In the algorithm, the segmentation on each scale can make full use of the information on all scales. The relationship between each wavelet coefficient can reflect the feature of each pixel location, and the interaction of neighborhoods can reflect the regional image segmentation. Both of these two fields constrain each other by product of joint probability, and act on the segmentation process together. The image segmentation procedure is sequentially executed from the coarsest resolution scale to the finest resolution scale, and segmentation result of the coarser scale is directly projected on the nearest finer resolution scale as its initial segmentation. The segmentation result on the finest resolution is used as the final result of the algorithm.