针对传统基于图论的图像分割方法在分割纹理图像或有突变边缘的图像时,分割效果不理想的问题,提出了一种新的基于图论的颜色一纹理区域分割方法,该方法采用分块的小波能量直方图与三阶颜色矩组成块特征向量生成图顶点的集合,同时根据块间距离进行加权,计算块间的相似度,作为图的边权值,再利用最小生成树的方法进行图像分割,方法有效地弥补了传统基于图论的分割方法中的两个问题。实验结果表明,该算法具有良好的分割准确性与鲁棒性。
Graph cutting is a popular technique for image segmentation. However, it is invalid for images with texture or strong edge. In this paper, we propose a novel segmentation method that graph-based color-texture region segmentation, which overcomes the shortcoming of existing graph-based segmentation method. This method computed feature vector of blocks using color moment, wavelets energy histogram and distance of blocks. Then we segment the image based on a minimal spanning tree. The experimental results show that our method is robust and efficient for image segmentations.