Application Of Dyadic Wavelet In Image Retrieval

The content based image retrieval(CBIR) is a hot topic mainly researched in the current computer vision field, and it is the growth and application of science and technology. In the condition of increasingly growing image database, image type diversification and sharply increasing quantity, the demand of quickly and accurately retrieving the images required by the user is gradually taken into account. This technology describes the content features of images using the shape, color, texture and other low-level visual characteristics of images, and then measures the similarity between images based on the characteristics of images, so as to achieve the image retrieval. Obviously, accurately extracting useful low-level visual characteristics of images is very important for CBIR; in recent years, the image retrieval technology based on a single feature has obtained some research achievements, but a single feature cannot completely express the image’s contents; therefore, combining two or more features becomes a possible way to improve the retrieval accuracy. However,how to effectively combine many different features to completely express the image’s contents in a mutually complementary way and retrieve the images suitable for visual perception by human eyes is a hot topic worthy of further research.According to the main low-level image features of vision, this paper makes a research in the matter of how to effectively extract image features, and utilizes the multi-scale feature of Wavelet Transform to achieve the improvement of feature extraction algorithm. The dyadic wavelet is a wavelet function that is between continuous wavelet and discrete wavelet and has a strong directionality; it has the translation invariance, can be used to avoid the visual deformation resulted from nonlinear transformation, and is equipped with the ability of noise resistance, which facilitates its universal application in the image feature extraction.In this paper, the shape, texture and color features are firstly extracted for image retrieval of a single feature; as for the color feature, the morphology principle is firstly used to detect the color edge of the image, and then the color feature is extracted from the color edge. For the texture feature, the dyadic wavelet filter with good nature and competence is selected, and then the process dyadic wavelet transformation and local binary pattern are adopted for extracting the image texture feature. For the shape feature, the binary wavelet modulus maximum edge detection method is adopted to extract the edge feature, and then the invariant moment is used to express the shape feature. Finally, the shape, texture and color are effectivelycombined for multi-feature image retrieval, and the simulated test is carried out in the MATLAB environment in recognized standard image database. The result of test indicates that the extracted feature vectors naturally include multiple low-level information, comprehensively reflecting the image contents; besides, it can be learned by comparing with the current similar algorithms that this algorithm has not only better precision and recall ratio, but also the strong robustness for illumination transformation and geometric transformation existing in the image.