关键词: 压缩域, 自动标记, 阈值最小生成树, 图像分割

Abstract: Due to the defect of image segmentation algorithm computational complexity in traditional pixel domain, this paper proposes a fast method for image segmentation in compressed domain. It segments images into blocks and marks image blocks automatically through Discrete Cosine Transform（DCT） coefficients and color moments in combination with Support Vector Machine（SVM） in compressed domain. It puts forward the Threshold Minimum Spanning Tree（TMST） algorithm to link marked blocks. It modifies and smooths edge of the segmented image according to morphology algorithm. The study validates the proposed method through Corel image database. The result indicates this algorithm is faster and more effective in image segmentation compared to previously proposed segmentation methods.

Key words: compressed domain, automatic markers, Threshold Minimum Spanning Tree（TMST）, image segmentation