Research on ROI extraction of color image

This paper studied a new method of ROI extraction based on quaternion matrix. Firstly, the image is cut into sub-blocks, and the Candidate region is gained by image entropy. Then the greatest expression of sub-blocks is obtained by svd of quaternion matrixs, and the relationship between the image blocks is found according to the definition of quaternion inner product. At last, dimensionality reduction of incidence matrix is carried out with Isomap, and in this process the ROI is extracted. The results of experiments confirm the feasibility and efficiency of this method.