Image Retrieval Technique By Color And Texture Feature Based On MPEG-7

In recent years, content-based image retrieval emerged as a very active area of development, the new technology is endless, but different CBIR systems usually have a different program of content description, this makes the retrieval of less common, and the reuse of resources is low. While the MPEG-7 standard formulation solved this problem, we know that the image feature extraction and search engine are not in the scope of the MPEG-7 standard, MPEG-7 is only made on the standardization of multimedia content description, and provides the common interface of other two parts, thus, the image feature extraction algorithm directly affects the performance of image retrieval. In this paper, a detailed study of the feature descriptor extraction and similarity matching algorithms which based on the MPEG-7 color and texture descriptors has been made and some improvements was proposed, the purpose is to further enhance content-based image retrieval of the versatility and efficiency.In the first two parts of the paper, the author studied the content of MPEG-7 standard and content-based image retrieval, and then searched the relationship between them in the current situation. After that, the author proposed a image retrieval system model based on MPEG-7 and summarized the commonly used similarity measure methods, normalized methods and evaluation criteria. At last, the test program used in this paper was also given.In image texture analysis, the author compared the MPEG-7 edge histogram descriptor with several classical edge detection operators, the comparative experiment shows that the direction information expression of the former is more obviously; The feature extraction algrithm was improved on the basis of the MPEG-7 edge histogram descriptor, that is, not only the edge of the original 80-dimensional descriptor is calculated but also the further description that contains the global edge and regional edge based on 80bins histogram is calculated, thus, we obtained the extended 100-dimensional histogram. Experiments show that the improved algorithm has the better search properties.In the analysis of the MPEG-7 color layout descriptor, extraction algorithm based on the color layout descriptor has been improved, experimental results show that the color distribution of the information is more obvious; For the dominant color descriptor extraction, the author analyzed the characteristics of human vision and used a 33bins dimension measure which is more reactive to human visual response in HSV color space, the comparison of experimental results was also given.After comparing the color layout and edge histogram of one image, the paper presents a new combined description that integrated color layout and edge histogram, this kind of visual decriptor is called regional color&texture density descriptor. Image retrieval tests proved that this group of discriptors is very simple and effective.Based on the above work, the author proposed a combination retrieval strategy, combination retrieval experiments show that a combination retrieval of visual characteristics has a better result than a single feature retrieval by allocating different weight ratio of main color, color layout and color&texture density.