Research On Shape-Based Image Retrieval

Shape is the low-level visual features of an image,then human visual system can quickly use the shape of the object to identify this object. The main object's image texture and color may be different performance characteristics, but the shape feature can be very similar indeed. Because the content retrieval of visual and semantic is not through the image of the atomic property or its key, so the shape of the image search feature is the basis of visual and semantic content. In this article, we offer three types of shape feature based image retrieval method and described their respective feature extraction, feature representation and feature matching. With plenty of image data is not to limit the applications point of view it independently to test the abstract shape feature search results.Firstly, to extract the main target of the image and background, the main object of the extraction of the shape of the object feature extraction is very important. Detailed study of the main method of extracting the object boundary, the main target of the image can be accurately extracted. Then, the extracted shape understanding and said that this is also a computer simulation of a major difficulty vision. This paper presents a new shape descriptor, curvature scale Fourier shape descriptors in the frequency domain characteristics of the boundary shape contour noise for good robustness. Fourier shape descriptors curvature scale with scale, rotation and translation invariant features, for the treatment of non-rigid object deformation is also a very good performance.Finally, the shape descriptors of similarity matching, in which the generic Fourier descriptor, curvature scale space descriptors and Fourier descriptors scale curvature similarity matching were carried out experiments, using recall and precision as Evaluation of effects tools. The three algorithms for image database retrieval based on shape features experiment. Image database used is the international standard MPEG-7 shape algorithm on the test images provided by the library. Test images which include the basic shape of the various non-rigid transformation. Experimental results show that the algorithm, this new form of feature extraction is very valuable.