The Object Contour-Based Image Retrieval Model

According to theory of human vision system, people tend to evaluate or understand image by its shape, namely shape-based image retrieval, which is an intuitive and useful way to make use of image's information. To find out the image shape's features which accord with perceptive trait of human eyes is not a simple task, therefore shape-based image retrieval is the one of most challenging problems within content-based image retrieval. Once different objects are extracted from images, how to describe object shapes in a proper and effective way becomes one of the difficult problems in image processing.We adopt the pairwise geometric characteristics, Relative Angle and Relative Position, to describe object contours, and apply a geometric histogram to save characteristics. As shown in the experiments this paired of geometric characteristics can meet the demand of a shape description of scale, translation and rotate invariant.On the basis of existing shape-based image retrieval methods, this thesis applies the above contour-descriptor method to image retrieval, and sets up the image indexing by object contours. Experimental results show that this method is effective.