The Research Of Multimedia Image Retrieval Based On Content

With the popularity of the rapid development of network technology and mobile digital devices, network multimedia content to show explosive growth in recent years and it has become a trend. A wide variety of multimedia content in terms of storage, transmission and analysis process will give us a challenge in the search technology area. People in the face of complex and dazzling multimedia messaging options, the lack of an effective search tool will be a very troublesome thing. Traditional text keyword-based retrieval model have been hard to keep up with the demand for multimedia retrieval, it is in the case, the content-based multimedia retrieval model, is very important for human development in the field of information retrieval significance.Content-based Multimedia Retrieval (MULTIMEDIA RETRIEVAL BASED ON CONTENT) popular speak is based on the multimedia content itself semantic retrieval, which is essentially different from text-based retrieval.Firstly,In this paper the multimedia data and retrieval models made a simple review summarizes the overview of the development of the model of image retrieval, video retrieval model, audio retrieval model. Details of multimedia content-based image retrieval model, and discusses the key technologies for the retrieval model launched. For example, the image feature extraction, image indexing techniques, relevance feedback algorithms.Then, starting from the image feature extraction, to discuss the same feature extraction algorithm based on SIFT image, the SIFT feature vector generation and matching process in-depth discussion and analysis, summarize the advantages and disadvantages of the image feature extraction. Finally, the lack of SIFT extraction process, such as scale space construction took, computer complexity and higher made a number of improvements. The SIFT algorithm has been improved, mainly through the scale space construction process optimization and dimensionality reduction of feature vectors, and take advantage of the improved SIFT algorithm for image retrieval model experiments, and have achieved good results.The main contribution of this paper is to:1) Analysis of the pros and cons of all major image feature extraction method, for the SIFT deficiencies, and some improvement;2) propose the SIFT-based image retrieval algorithms based on the improved, the algorithm uses the Gaussian kernel size of the adaptive method to build the pyramid image, reduce the computational complexity, and then follow the traditional steps of feature extraction, with the Euclidean distance to measure similarity, and finally the use the BBF search algorithm for image retrieval. The preliminary experimental results show that the image retrieval model is able to detect the target image after translation, scaling, affine, illumination changes, It is quite applicable to typical image retrieval both in exacting time and efficiency.