Image Feature Extraction Technology Based On Local Binary Pattern

In image retrieval and related image processing techniques,the feature extraction of images occupies an important position.The effectiveness of extraction features and the speed of extraction directly affect the performance of subsequent analysis.Since the local binary pattern has the characteristics of high efficiency and fastness,it has been widely applied in face recognition,image texture classification and image retrieval.Based on the research of the existing local binary pattern algorithm,this paper mainly analyzes and summarizes its algorithm and its related improved algorithms.In the light of some limitations and shortcomings of the existing algorithms,it proposes relevant improvement strategies.The main contents of this paper are as follows:(1)Make some concrete analysis and research on the current popular local binary pattern algorithm and its related improved algorithms,and summarize and put forward some characteristics and limitations of the existing algorithms.(2)Aiming at the problem that the local Mesh Quantized Extrema Pattern(LMEQEP)can not extract the difference information between pixels and extract the feature dimension too large in the texture feature extraction process.In this paper,an optimization algorithm is proposed.Firstly,the idea of using LMEQEP operator to extract the extreme value information around the central pixel is combined with the improved extreme value information resolution strategy.The improved extraction strategy can effectively avoid the algorithm is too sensitive to the change between pixels and improves the robustness of the algorithm.Secondly,on the basis of the improvement,in order to avoid the dimension explosion caused by too long coding length,the circularly symmetric coding method is adopted to classify the extracted pattern coding.The improved method can effectively reduce the feature dimension.The new algorithm greatly improves the retrieval ability of the two commonly used image databases,and achieves the goal of desensitization and dimension reduction at the same time.(3)The calculation of the classical local binary pattern and its related improved algorithms are based on the correlation between the central pixel value and the neighboring pixel value,and encode the quantized correlation,and finally to represent the image Characteristics.In order to further deepen and popularize this method,an extended local binary pattern algorithm is proposed.Firstly,the robust gradient image gradient information is used instead of the original pixel.Secondly,the existing correlation calculation between pixels is extended to the correlation between gradient regions.This can effectively analyze more information about the correlation between image regions and avoid the disturbance caused by the slight change of local pixels,which makes the algorithm have better robustness.Finally,in order to more subdivide this correlation,the quantification of the algorithm is expanded from the original binary quantization to the five-valued quantization,and the method of splitting is used to reduce the coding calculation of the algorithm.In two commonly used image database retrieval experiments,the optimization algorithm has better retrieval performance than other local binary pattern algorithms.