Research On ROI Extraction

With the quick development of multimedia, Internet and image processing technology, the retrieval of image database is becoming an urgent problem in the field of information retrieval. Because the users'need is the most important content in an image retrieval system, it is necessary to conduct a user-friendly image system. In view of image retrieval based on the region of interest (ROI) is close to the users'need, it has been becoming the hot topic in the field of image retrieval.The key step of image retrieval based on ROI is how to extract the region of interest. Due to the uncertainty of the image content and the imperfection of image segmentation, there is not a common method to get the ROI of an image until now. Based on the review of the concept and characteristics of ROI and the current research status, three novel ROI extraction methods are proposed in this paper.(1) A ROI extraction algorithm based on edge detection and RS theory is introduced. Firstly, rough ROIs are determined based on the priori-knowledge. Then, the relevant low-level features, such as intensity, edge, location and so on, are extracted and an information table is constructed. Based on the discretization of the information table, the basic regions are built. Finally, the ROIs are taken as the upper approximation of original ROIs.(2) A ROI extraction algorithm based on conventional region growth algorithm and RS theory is presented. Because it has not specific conditions or rules to restrict region growth in conventional region growth algorithm, RS theory is recommended in the paper. Firstly, a knowledge representation system is constructed by the saliency, the relative position and color information of the saliency image. Then, a useful region growth decision-making rule is derived by attribute reduction and value reduction. Finally, the ROIs are extracted based on the decision rules. The new method reduces the limitations of the traditional region growth algorithm and improves the accuracy of ROI extraction.(3) A ROI extraction algorithm based on point of interest is presented. Firstly, the corners of an image are detected based on multi-scale curvature polynomial. Then, the minimal convex polygon of the corners is introduced as a ROI of the image. The new method is not affected by image rotation, scaling and affine transformations.