Research On Image Retrieval Based On Nonsubsampled Shearlet Transform And RI-LPQ

With the rapid development of Internet, communication technology andstorage technology, a lot of pictures emerge. How to retrieve the requiredimage efficiently and accurately from a large amount of images is an importantissue in image analysis and application fields. Content-based image retrievaltechnology when retrieve huge amounts of data is a very effective method.Thetexture is one of the key elements characteristic description of the image.Wavelet transform has good time-frequency analysis capabilities, It has been widelyused in image retrieval. But the traditional wavelet transform has limited directionselectivity to capture edge features and cannot effectively express the geometriccharacteristics of the edge of the image. In order to flexible to capture the directioninformation of texture image, the method of multi-scale geometric which has directionselectivity and anisotropy developed rapidly. Nonsubsampled Shearlet Transform(NSST) exhibits highly directional sensitivity and shift invariance， even it can besparse representation of the image. In contrast， NSST acquires the natural texture andedge information with the traditional wavelet and has higher computationalefficiently with Contourlet. Ojansivu proposed rotation invariant local phasequantization description operator which has a strong effect texture description and hasbeen successfully applied to face recognition and image classification. This article willbe used for texture image retrieval.This thesis studies the relevant properties of NSST and its application in imageretrieval.It firstly acquires the statistical features of the image NSST coefficients byGeneralized Gaussian Distribution Function. Then, image features are directlyextracted by Rotation Invariant Local Phase Quantization description operator. Finally,texture images on the Brodatz image database are retrieved by the formula of similaritymeasure with weight coefficients. The experiment result indicates that compared thismethod with traditional wavelet and Contourlet, the former obtains better retrieval rate.