ROI Selection and Image Retrieval Method Based on Contribution Matrix of SURF Features

In traditional image retrieval method, features need to be extracted within the whole region of image, which leads to high computation and semantic ambiguity. To address this issue, this paper proposes a technique to select region of interests（ROI）, and carries out the retrieve process within the ROI. Firstly, SURF feature descriptor is used to extract local features and keypoints. Then, dynamic program is employed to calculate the sum of sub-matrix of feature points distribution, which is finally utilized to extract the ROI. Finally, we integrate the color, texture and shape features into a fused feature within ROI, and use nonlinear Gaussian distance function to retrieve images from the database with user input. Experimental results show that our proposed method has high conformity with human vision, and is effective for image retrieval.