Research On Cultural Relics Images Classification Based On Multi-features Fusion

In recent years,with the increasing awareness of the protection of cultural relics in the whole society,the research on the protection of cultural relics based on computer-aided technology has been widely concerned by researchers.As an important part of the research on computer-aided cultural relic protection,the technology of image classification has important application prospects in the management,protection and development of cultural relics.Based on the study of image classification algorithm,the research on the key technology of cultural relic image classification is carried out by using the established cultural relic image database and the feature extraction and classification algorithm for cultural relic image is realized.the main research results are as follows:1.In this paper,a total of 700 pieces of cultural relics were collected,through the network and field acquisition,and the database of the cultural relic image was established,which lays the foundation for the research of this paper.2.Image feature extraction.Based on the analysis and research on the characteristic of cultural relic image and commonly used visual features,this paper uses HSV color space extract color histogram feature of cultural relic image;texture features of cultural relic image is extract by Gabor filter;local feature information of cultural relic image is extracted by using the characteristic of scale,rotation and illuminatio n invariant of SIFT and HOG;The four characteristics extracted from this paper describe the characteristic information of cultural relic image from the global,local and othe r aspects.3.Feature fusion.In order to realize the complementary advantages between the single feature,a weighted multi feature fusion method is proposed based on the classification accuracy of single feature to decide the weight,at the same time,based on the research of the method of weight calculation,in this paper,a new method is defined to select the weight of each kind of image based on the distance of the image,by calculating the sum of products of accuracy and corresponding weights of all kinds of images based on the different features,the weighted multi feature decision fusion is realized.4.Classifier design.In response to the error accumulation of the traditional binary tree SVM classification,this paper proposes an improved SVM classifier based on binary tree.First,the fuzzy C mean clustering algorithm is used to accurately determine the types of artifacts in the image center,then the similarity between different classes is calculated,and the classification order of different classes is determined according to the similarity between categories.Based on the above research methods of feature extraction and classification,this paper implements the image classification algorithm,and verifies the effectiveness of the algorithm.