Research Of Design Patent Images Classification

With the knowledge-economy emerges vigorously, countries all over the world pay more and more attention to the intellectual property protection, development level of patent has been a strategic sign of measuring the overall strength, development capacity and core competitiveness of a region. Therefore, it has a profound and significant social and economic significance to sharpen the enterprise product innovation ability and reduce the profession patent infringement dispute, which to accelerate the development of patent retrieval technology and improve the efficiency of patent retrieval.Content-Based Image Retrieval(CBIR)extracts visual features as retrieval features, such as color, texture and shape, etc. For the existence of semantic gap between low-level image features and human understanding to images, CBIR can't get satisfied retrieval results. Classifying images into reasonable categories using low-level features or annotating images will greatly improve the performance of CBIR systems.This thesis does a study of image classification based on low-level image features. We use the method of bounding box-contour distances (BBCD features) of appearance design patent image combined with block features to extract the low-lever visual features at first, and propose the algorithm of classify index based on K-means Clustering, to organize the index structure of data in patent image database considering the semantic similitude and the visual features similitude, so we could realize effective classification and save time for the follow-up retrieval in the same time. Classification Concept based on K-means meets the characteristics that there are big differences in different categories and something similar in the same category. It may reach a good classification result in small Design Patent Library, but there are some limitations in K-means, such as the influence of the initialization of cluster centers. Compared with the learning methods based on empirical risk minimization, SVM presents a lot of especial advantages for resolving the problems with nonlinear as well as small samples, and has no problem of immersing into the partial least point, so Image Classification based on SVM can make up for the inadequacy of the k-means method. We studied the principle of SVM, constructed Multi-category SVM classification, and used optimizing method of parameter to realize the Image Classification based on SVM.