The Research Of Shape Representation And Matchingalgorithm Of Image Retrieval

With the development of network technology and the popularity of digital imaging equipment, large-capacity image databases have wide application in all walks of life. How to find the relevant image from an image library quickly and effectively is a challenging task. As a result, content-based image retrieval applies for us. Content-based image retrieval is no longer rely on manual tagging for text retrieval, but rely on the image inherent in the color, texture, shape and spatial relations and other features to retrieve similar images. Using shape descriptors to depict image has great superiority, such as the shape features are often closely linked with the target object and the object's shape does not change with changes in the surrounding environment. Therefore, the shape-based image retrieval becomes a research hotspot. How to find effective ways of describing the shape and shape similarity calculation method is the core issue. Main contents and contributions of this article are as follows:(1) Several key techniques and algorithms of CBIR are deeply analyzed and discussed, especially for shape representation and shape matching methods in shape-based image retrieval. The bottom visual features of the image such as color, texture, shape and spatial relation has been done a more detailed description. In addition, the similarity measure between images'feature and the evaluation criterions for image retrieval algorithms were also introduced correspondingly.(2) A shape representation and matching method based on edge gradient orientation statistical code (EGOSC) is proposed. Firstly, constructing the 18-direction vector and making maximal summation restriction on EGOSC to make sure this algorithm is invariable for rotation effectively. Secondly, we apply EGOSC in shape-based image retrieval, and finally a corresponding shape matching method is proposed. We use the edge gradient orientation entropy (EGOE)'Euclidean distance to measure similarity of images, so that this method is not sensitive to scaling, color and illumination change. Experiment results and algorithm analysis demonstrate the efficiency and feasibility of this shape-based image retrieval approach.(3) A new shape representation and retrieval method called distance autocorrelogram is introduced. Distance autocorrelogram are obtained under the premise of getting the contour'centroidal distances. Then, we apply this shape descriptor to CBIR. This feature depends on the centroidal distances and correlation between neighboring edges. So, it can express the edge'spatial distributing informations. This scheme is effective and robustly tolerates translation, scaling, rotation. Experiment results and algorithm analysis demonstrate the efficiency and feasibility of this shape-based image retrieval approach.(4) An improved distance coherence vector for shape representation and matching is proposed. in order to improve the algorithm proposed by Sajjanhar. Our improved method regards centroidal distances of average coordinates which are from the biggest connected domain in coherence pixels bins as a new feature vector. The new added vector is invariable to translation, scaling and rotation. Similarity of images is measured by different similar functions according to different feature vectors. This new algorithm has made better retrieval effect because of the more spatial information introduced.