Research Of Shape-Based Features And Matching Algorithm In Image Retrieval

With the rapid development of multimedia technology and network technology, digital image information is increasingly widespread in all aspects of life. How to retrieve the needed information from the large amount of image information quickly and accurately is becoming an urgent problem. The shape-based image retrieval is an important part of the content-based image retrieval. As one of the essential characteristics of the object, shape carries a certain semantic information, other than the color, texture, and spatial position relationship. Based on the shape of the image retrieval is also a difficulty of content-based image retrieval. Shape-based image retrieval has a particular significance, which is the result of a combination of multi-disciplinary. With the improving requirements of the information management, shape-based image retrieval is used in a wider boundary. Processing shape information is mainly to use some part or whole of the image features, and use some classification comparison. Expression and similarity judgment method could have a lot of very different approach. This article makes a detailed description from the global features of the image description and partial characterization, analyses the difference between two different kinds of image description, and select some typical algorithms for analysis and improvement.The main work of this article is as follows:(1) Analyze the commonly used global feature extraction algorithm. We make a detailed description of the wavelet algorithms and implementation steps. Also we descript another global features image description algorithm--HOG algorithm and its implementation steps. Based on these algorithms, we build a retrieval system which achieves a good performance.(2) This paper analyzes a recently proposed partial feature extraction algorithm--local self-similarity description and its implementation steps. We represent a retrieval system based on self-similarity descriptors using bag of features ideology. Experimental results in the ETHZ data sets show the effectiveness of the proposed system.