Research On Content-based Scanned-certification Image Retrieval Method

Abstract:With the development of computer technology and network technology, scanned-certifications images are attached to various kinds of rewards applications or corporation’expand brochures as important proof materials. In order to guarantee the legal use of those images, avoid repetitive usage of the same certification, duplicate checking for scanned-certification images in some special data sets is very important. Though there are an amount of image retrieval systems for various kinds of image data sets, relatively little attention is drawn on scanned-certification image retrieval. Therefore, the research on the content based scanned-certification image retrieval methods has important theoretical significance and application value.In this paper, the basic theory of the content-based image retrieval are first expounded, along with current research and application situation. Then some key technologies of content-based image retrieval are introduced. Due to the poor image quality and slant position problem, image preprocessing methods are indispensible, including image denoising achieved by Gaussian filter and an arbitrary skew correction method for scanned-certification images. Considering the specialty of certification images, we proposed an image retrieval method based on local object for scanned-certification images, where the round seal is the local object. The experimental results show that, compared with the retrieval method based on single color histogram, this method has better performance. On the basis of the above study, the random forest algorithm is introduced to further improve the retrieval performance of scanned-certification images. On the basis of above scanned-certification images retrieval method, the new retrieval framework based on random forest take advantage of random forest’s superior classification performance, the experimental results will prove the effectiveness and superiority of the retrieval framework.Based on the scanned-certification image’characteristics and specific retrieval needs, this paper designs and realizes a scanned-certification image retrieval test platform with VS2008and open source library OpenCV2.1, on which the proposed image retrieval method was tested and analyzed. The experimental results show that the retrieval system constructed by this paper can effectively realize similarity retrieval of scanned-certification image, and shows good retrieval performance at the same time. The retrieval system runs stability and has friendly operation interface.