The Research Of Content-based Image Retrieval And Implementation Based On Embedded System

With the rapid development of the multimedia technology and industrial vision, the research about embedded image retrieval becomes a hot research at home and abroad. The establishment of an effective image description and retrieval mechanism based on various embedded platforms has become an urgent problem to solve. As a result, the technology of Content-based image retrieval was brought to light. Based on the research and analysis of Content-based image retrieval technology,aiming at removing the defects and limits of existing technics,this article explores several important problems in the common technics of Content-based image retrieval search systematically and profoundly and proposes an entire solution which is proved to be feasible through the implementation based on embedded system and through large amounts of experiments.Main researches of this article are:On the aspect of the image feature extraction, low level features like color,shape,texture and spatial relation and high level characteristics like semantics of image are analysed and researched in detail.And on this basis,several optimization strategies are proposed: in terms of the feature extraction based on color,a self-organised clustering method is adopted to optimize the Color quantization process.a table-searching method of weighted part of color is adopted to acquire the spatial information of the pixel. In terms of the feature extraction based on shape , Hu's moments invariants algorithm based on the vague-C average clustering splitting and Canny-T algorithm based on grey threshold value are proposed.Through experiments and analysis the proper situation for the two optimization methods is determined.On the aspect of system realization,a Content-based image retrieval embedded experimental system is realized.The system realizes both the traditional way of retrieval such as based on color(color histogram,accumulative histogram),shape(edge histogram based on Sobel,Canny edge detection) and the optimized way of retrieval mentioned in this article.The experiments prove that the optimized method mentioned in this article can increase the correctness and completeness of image retrieval.And it bears a theoretical and practical meaning.