The Research Of Design Patent Image Segmentation And Feature Extraction

Along with the design patent innovation in the role of competition, more and more attention paid to the protection of appearance design of growing importance. "Huge amounts of information "is the characteristics of patent database, therefore, to develop an automatic design pattern image retrieval and analysis system is of much necessity.In the design patent image retrieval system, image preprocessing and image feature extraction are two crucial parts of it. Design patent image segmentation is the basic work before retrieving and analyzing, and feature extraction is the core technology of image retrieval.Aiming at the properties of image, whose image background is diversity and complexity in the design patent database, some important problems related to design patent image segmentation and retrieval have been discussed in this paper. A multi-feature design patent image retrieval algorithm based on data fusion has been proposed, something about design patent image segmentation, image feature extraction and data fusion has been mainly studied. A whole set of experiment scheme have been designed. Experiments show that this scheme can effectively improve the efficiency of design patent image retrieval system.On the basis of brief introduction of design patent image retrieval algorithms home and abroad, the main work of this paper can be summarized as the following three aspects:First, the design patent image has the characteristic of diversity and complexity. Aiming at it, any single image segmentation algorithm can hardly meet needs. An improved algorithm of Canny operator combined with threshold segmentation is introduced, which can obtain the threshold according to the sampling and much fit for the various images in the design patent image.Second, feature selection is the key point of image retrieval, which need to combine image feature with retrieval demand. For any kind of features can hardly express the content of design patent image, two kinds of essential attribute——shape and texture, have been chosed as the feature set for retrieval. An algorithm called Histograms of Gradients(HOG), which was applied to object detection originally, has been used in the feature extraction for image retrieval. At the same time, a simple texture descriptor has been put forward. The algorithm process and experiment contrast have been given.Third, there is a contradiction between real-time and retrieval efficiency in the retrieval system, and how to get the compromise is a problem considered for all the retrieval systems. The dimension of HOG has been decreased by principal component analysis(PCA).On the basis of different importance grade, multi-dimensional features have been integrated with weighed fusion, which effectively keep important characteristics' information and shorten retrieval time simultaneously.