The Reasearch Of Image Retrieval Based On Gabor Wavelet And Adaptive Local Phase Quantization

With the rapid development of network, computer and multimedia technology, more and more information is transmitted and saved by image as the medium. For a large number of those image data appearing in working and living, How to accurately and quickly extract the image features and retrieve the target image has become the forefront of intelligent information processing issues, and has practical application value.This article analyzes the research status and methods of image retrieval for image feature extraction; feature extraction method is proposed and used to realize image retrieval. In the image feature extraction, the theoretical basis and characteristics of Garbor wavelet, Binary Pattern(local),(LBP) texture feature, local phase mode(Phase Quantization Local, and LPQ) are analyzed. After analyzing the existing shortcomings of the Garbor wavelet and local two value pattern feature extraction methods, a method based on Garbor wavelet and adaptive weighted local phase quantization is proposed in this paper. The main steps are as follows, firstly, designing Garbor filters group with a scale of 5 and the direction of 8, the filter is used to process the image, and a total of 40 filtered images can be obtained. Secondly, the 40 images are divided into sub images of different size pixel blocks, and the texture features are extracted by using adaptive weighted local phase quantization method for each sub image in different sub blocks. Finally, according to the information entropy of each sub block, the sizes of the information entropy are given different weights and add to the image texture features. The feature extraction method in this paper integrated the noise reduction performance of multi-scale wavelet decomposition of the Garbor method and the advantage of local phase quantization texture not sensitive to the illumination transformation. Mahalanobis distance calculation characteristic similarity is used in image retrieval similarity measure.In this paper, the software of Matlab is used for image retrieval methods and compare to other methods. The results show that the method proposed in this paper has a better precision in image retrieval under the same conditions.