Research On Local Pattern Descriptor In Image Recognition And Retrieval

With the rapid development of the Internet,the data of image media is growing exponentially.It is a problem that people pay great attention to find the required image from mass data.The quality of feature extraction and description directly affects the final performance of the task.How to extract high-quality image features from images with large noise interference,complex background and large change of target posture gives a challenge for researchers to recognize and retrieve target images quickly,efficiently and accurately.Because local feature extraction and description are often important steps to find target images,the in-depth study of local pattern descriptor has important theoretical and practical application value.This thesis focuses on the methods of extracting the underlying features of images,especially texture features,and applies the proposed local pattern feature descriptor to the recognition and retrieval of texture images and natural scene images.This thesis mainly completes the following research work:(1)Based on the original local binary pattern,the existing improved descriptors are studied,including uniform rotation invariant local binary pattern,local pattern Fourier feature,double local binary pattern,extended coding local ternary pattern and scale invariant local ternary pattern,and so on.Support vector machine is used for image recognition and cross discriminant quadratic analysis is used for image retrieval.The experimental results show that the robustness of the above descriptors for images with rotation changes needs to be further improved.(2)An Anti-rotation scale invariant local ternary pattern descriptor is proposed.Firstly,the image is grayed out,and the sample texture feature is extracted by using the scale invariant local ternary pattern descriptor.Then,according to the rotation invariant mapping mechanism in the rotation invariant local binary pattern,a mapping pool based on the scale invariant local ternary pattern is constructed.Finally,according to the anti-rotation mapping rule and histogram calculation,the anti-rotation scale invariant local ternary pattern feature values are obtained.The experimental results show that the proposed anti-rotation scale invariant local ternary pattern descriptor can effectively adapt to image rotation changes.(3)Combining the anti-rotation scale invariant local ternary pattern with color features,the recognition and retrieval of natural scene image and texture image are realized.First of all,the texture feature of the target image is extracted by the anti-rotation scale invariant local ternary pattern,and the color feature based on HSV color space is extracted.Then the serial fusion of these two features is realized.Finally,the image is recognized by support vector machine classifier and retrieved by cross quadratic discriminant analysis.The experimental results show that the effect of image recognition and retrieval based on fusion features is better than that of single local feature.