Research On Image Retrieval Method Based On Multi Features Fusion And Its Application

Today, three important branches have been formed in image retrieval field, which are text based image retrieval, content-based image retrieval, and semantic based image retrieval. Text based image retrieval uses text to describe the user’s needs, such as image name, the image features. Because text expression ability is limited and text tagging is Ambiguous in the process, text based image retrieval often does not meet user requirements. Semantic retrieval, refined the high-level semantic expression ability on the basis of image visual features further. But its retrieval process is complex and the technology system development is not perfect. Content based image retrieval, in color, texture, shape, etc. as the image feature expression and similar judgment, is the research hotspot in the field of image retrieval.In order to improve the precision and recall, variety of features need be used in content based image retrieval method. The fusion strategies and rules of these features are the key issues in the retrieval performance of content-based image retrieval methods.In this doctoral thesis, content based image retrieval method is taken as the research object, and multi feature fusion framework and fusion strategies are taken as core research contents. Taking detection field as research background, research work are carried out as follows:First, multi feature fusion of image retrieval theory had been researched based on single feature image retrieval framework. They had been analyzed such as feature selection, feature extraction, feature vector generation, and similar comparison. A general framework for the theory of multi feature fusion image retrieval is constructed from two aspects of feature vector fusion and similarity measure fusion. Based on the analytic hierarchy process, the evaluation framework of image retrieval algorithm is constructed.Second, an image retrieval algorithm based on color feature and edge feature fusion is proposed in order to satisfy precision and fast requirement. In the similarity measure, R, G, B color features and edge feature is fused together, in order to enhance the accuracy of the retrieval results. At the same time, the extraction of color feature and edge feature are put on the low frequency component image of the two-level wavelet decomposition in order to reduce the execution time of the retrieval process. The experimental results of performance evaluation show that the algorithm can meet the requirement of precision and rapidity.Third, an image retrieval algorithm based on the characteristics of wavelet bases is proposed by considering recall and rapidity index. This algorithm fused two classes of wavelet feature as similarity judgment of the image retrieval process which give full play to the adaptive ability of the wavelet feature and make the algorithm have strong pertinence to different categories of image. At the same time, the characteristic of wavelet bases is replaced by Taylor series approximation, and it also has a good speed advantage.Fourth, the image retrieval algorithm based on multi feature ESN fusion is proposed by considering the precision and reliability index. This algorithm fused color feature, texture feature, and shape feature, and gained three feature fusion weights of ESN training of the network, which not only enhanced the comprehensive similarity judgment also enhanced the adaptability of the retrieval process.