Image Retrival Method With Multi-feature Based On Dempster-shafer Theory

Content-based image retrieval means of retrieving and querying images byimage content. The traditional content-based image retrieval method mainly used thesingle features of the images based on color, texture, shape features, or multi-featurebased on simple weighted. The retrieval rate of existing methods is still not high, inorder to improve the retrieval accuracy, this paper proposes an image retrievalmethod with multi-feature based Dempster-Shafer theory. The main research workare as follows:First, this paper analyzes the existing method of extracting color features, thenputs forward the image retrieval algorithm combined with multi-color features. Atpresent, the single color feature cannot describe the color informationcomprehensively and accurately. For example, the most commonly used colorhistogram in the image retrieval, although it can be described the rough proportion ofvarious colors in images, it cannot describe the relative position of different colors.The color moment with more comprehensive information cannot express image colorspace position yet. Therefore, in view of the above problems, this paper puts forwardthe image retrieval algorithm combined with multi-color features, the methodcombines the color histogram, color moments, and color correlation of expressingimage color space position. The results show that the retrieval algorithm accuracy ishigher than the retrieval algorithm only based on the single color feature. Therefore,this paper uses multi-featured fusion of color histogram, color moment and colorcorrelation.For the image texture feature extraction, this paper analyses the characteristicsof the current texture features, and compares the commonly used Gray level co-occurrence matrix, Tamura texture features and Gabor wavelet transform through the retrieval experiment. The results show that the image retrieval method based onmulti-scale Gabor wavelet transform are better than the other methods. Therefore,this paper chooses the multi-scale Gabor wavelet transform which has good spatiallocality and direction selectivity as texture features. This paper also analyses andcompares the applicable conditions of similarity measurement method throughexperiment, and uses the query result accuracy of maximum Euclidean distance tomeasure the similarity between images.Finally, this paper uses the color and texture feature fusion, introduces theDempster-Shafer theory, and puts forward an image retrieval method with multi-feature based on Dempster-Shafer theory; Then design the image retrieval systembased on multi-featured fusion, The results show that the image retrieval algorithmwith multi-featured fusion can effectively improve the accuracy of image retrieval,and meet the actual needs of content-based image retrieval.