Textural Image Segmentation Using Discrete Cosine Transform

In this paper we propose a computational efficient approach for image segmentation based on texture analysis, a 2D discrete cosine transform (DCT) is utilized to extract texture features in each image block. We first split the input image into MxN blocks, calculate the distances between neighbor blocks by a set of largest energy signatures from DCT for each block. Then we merge blocks with smallest distances to form larger regions. The process will repeat until we got desired number of regions. Experimental results show that our proposed method outperforms the existing image segmentation method, especially on efficiency aspect.