Research On Image Retrieval Technology Based On Intelligent Optimization Algorithm

With the progress of science and technology and the development of the Internet era, as well as the advent of the era of big data and the extensive use of multimedia technology, makes the image data show a trend of explosive growth. How to accurately retrieve the desired image from the vast image database became a hot research topic in the field of computer in the past several decades. In order to achieve the accurate image retrieval, the use of image content, such as color, texture and shape features to retrieve image technology comes with the tide of fashion. This technique is called Content Based Image Retrieval, which has broad application prospects and far-reaching research value and business value. So it is paid great attention by the related researchers and research institute. Although the CBIR technology already has achieved good results and some has been used widely, there are still many aspects that need to be further improved and optimized.Due to the method of the traditional Content Based Image Retrieval using single visual features and similarity measure algorithm has limitation, this paper propose and implement one Content Based Image Retrieval method that use two kinds of visual features, color and texture as well as twelve kinds of visual features similarity measure algorithm and use QPSO to optimize it. At the same time comparing the search result of the four algorithms, one most optimal algorithm is derived. Then use GPU acceleration technology to improve search efficiency. Key technical and theoretical methods used herein include the following four aspects:(1) In this paper, the integrated use of two kinds of visual features,image color and texture achieve Content Based Image Retrieval. For color characteristics, based on human visual characteristics use four kinds of color space, RGB, HSV, Lab and Gray to extract image color histogram and color moment feature and quantify them. For texture characteristics, use two main methods of image feature description, GLMC and Gabor to extract image texture characteristics and quantify them.(2) Use the current commonly used twelve kinds of similarity distance algorithm to measure the color characteristics and texture characteristics between the target image and each image in the image library.(3) Use four kinds of Swarm Optimization Algorithm, PSO,QPSO, CLPSO and SLPSO to acquire the approximately best combination of the optimized features, similarity measure function and measure weight, which make the retrieval more accurate and more efficient.(4) For the best one of the four kinds of Swarm Optimization Algorithm, QPSO, use C++ AMP technology to implement GPU acceleration to the system and verify the effect of the acceleration by testing.