Key Technologies On Image Pre-Processing In Design Patent Retrieval

New emerging Content-based image retrieval is putting forward for design patent,because the traditional text-based retrieval system cannot achieve good performance onsuch data that is dominated by images. In essence, the retrieval is taken on foregroundobject, for the foreground composes the primary protection object instead of thebackground. Without taking account of this special attributes, the general purposecontent-based image retrieval system cannot reach the demanding precision on patentimages. From the analysis above, it is easily to note that preprocessing of images is anelementary part in patent retrieval.According to the difficulties encountered in image preprocessing, an effectivepreprocessing measure is proposed for the different fields of design patents, whichcontributes a lot to the query precision. This method mainly has three folds as follows.Firstly, for the diversity of users’ upload images which may suffer from heavy clutteringbackground, Interactive segmented method is proposed to locate the interest region ofimages. Secondly, for lack of standard, batch foreground-cutting algorithm is proposed totackle the problem caused by different backgrounds of images in patents databases. Lastly,a character recognition algorithm is proposed to identify and extract the characters whichare frond and stereo view in patent images.Based on the algorithm above, foreground extracting, elimination and recognition ofcharacter in batch mode are implemented as well as interactive preprocessing in the WEBmode. According to evaluation on database, the performance of the retrieval system isimproved significantly, moreover, the rate of extracting front and stereo view come up to98%or more.