一种基于多字典学习的图像分割模糊方法

An Image Segmentation Fuzzy Method Based on Multi-Dictionary Learning

本文提出一种基于多字典学习的图像分割模糊模型和算法.在模型中,结合多字典学习和模糊方法,考虑了分割区域内部的一致性和边界的正则性:一方面使用区域块均值和带有类标的结构字典重构图像块,利用重构误差和l2正则能量共同度量分割区域内部的一致性,该度量能够刻画图像不同区域的灰度信息和纹理模式;另一方面采用小波系数稀疏正则保持分割区域边界的几何结构.基于交替方向乘子法和字典学习方法给出新模型的快速求解算法.在该算法中,除了小波阈值,每一步都是显示表达式,因此简单易用.一系列实验结果验证了本文算法的有效性.

This paper presents an image segmentation fuzzy model and algorithm based on multi-dictionary learning.In the proposed model the conformity within the segmented regions and the regularization of the boundary are considered by combining multi-dictionary learning and fuzzy method. On one hand,image patches are reconstructed by using the block means within the segmented regions and a structured dictionary with class labels. The class-specific reconstruction residual and the l2 regularization term measure the conformity within the segmented regions. This measurement can describe intensity information and texture pattern for different regions of images. On the other hand,the wavelet sparsity regularization is employed to preserve geometric shape of the segmented regions. Based on the alternating direction method of multipliers and dictionary learning method,we design a fast alternative iteration algorithm to solve the proposed model. In the proposed algorithm each step except wavelet shrinkage is a closed form. Hence it is easy to use. Numerical experiments are presented to demonstrate the efficient performance of the proposed algorithm.