Research On Feature Extraction Method Of Cultural Relic Image Based On LBP

The museum is a temple to display the excellent traditional Chinese culture,which is full of rich Chinese excellent cultural genes.The display of cultural relics in museums takes various forms.However,the cultural relic image recognition technology still relies on the traditional LBP algorithm to a large extent.The gray value of the center point is taken as the benchmark,and the gray value of the neighboring pixel points is compared with it in turn to code the image,and then the texture features are obtained.The traditional algorithm is not only time-consuming and laborious,but also has poor reliability and high false detection rate.Therefore,it is an urgent problem to improve the existing feature extraction technology.In this paper,two improvements are made to the shortcomings of the original Local Binary Pattern(LBP):(1)Two kinds of threshold values are set according to the different neighborhood points,and the LBP coding values are calculated respectively.The direction information between neighborhood points is added on the basis of the original features.(2)Convert the color component values of two color Spaces,RGB and HSV,extract the values of four color channels of image R,G,B and V,and carry out four-value normalization processing.The difference between neighborhood point and center point in two different color channels is represented by a string of binary,and then the color features are fused.The main research contents of this paper are as follows:(1)Summing up the related technologies of cultural relic image recognition,summarizing the concept,classification and development status of the original LBP algorithm,and understanding its application in scientific research and life.(2)In view of the shortcomings of the traditional LBP algorithm,two improvements are proposed: color feature extraction of cultural relic images is added;normalized image color processing is carried out through RGB and HSV color space transformation to facilitate subsequent feature extraction;The LBP coding values of the four horizontal and vertical neighborhood points were calculated based on the threshold values of their adjacent points by adding the directional features between the neighborhoods.(3)The data set of this paper is derived from eight types of cultural relic images of Datong Museum: jade,pottery,porcelain,bronze,painting and calligraphy,epitaph,rock painting and sculpture,with a total of 1040 cultural relic images.(4)This paper selects two methods for effect evaluation.The first is to use SVM classifier to classify and recognize cultural relic images and evaluate their effects.The second method is based on Matlab platform.The extraction methods before and after the improvement are applied respectively to calculate the feature vectors.The confusion matrix is calculated and visualized by Matplotlib function,and the classification effect is counted.