Book Recognition Algorithm In Point-Reader Based On Video Image

With the rapid development of modern digital IT, point-reader and othersupplementary products have been essential learning tool for most students. In order toaccomplish selecting books intelligently and promptly, with the fund of ShenzhenHuapu Electronics Technology Co.,Ltd., on the hardware platform of point-readingsystem built by our seminar, the book image recognition method have beening researchdeeply. Moreover, the main work and the results are as follows：(1) Study the methods based on the book cover image of USB camera acquisitionand preprocessing. Take advantage of VFW technology to build real-time videoacquisition and processing platform, which are endowed with video capture, videoediting, video playback and other functions. Meanwhile, For the noise in the imageacquisition process, we have studied image illumination compensation, gray transform,and Gaussian filtering preprocessing algorithms to overcome the noise disturbance andimprove image quality.(2) Study the method based on coarse matching method based on the books’ widthheight ratio. Firstly, extract the book cover border by the largest connected domaindetection as well as the least squares straight line fitting. Then, utilize the books aspectratio screening method for coarse matching. Therefore, select eligible books from thestacks. In this way, it reduces the subsequent computation amount to improve thematching efficiency.(3) Study matching method based on the book cover image feature pointsmatching. Firstly, a adaptive Harris corner detector algorithm is applied to extractfeature points from the test book cover map and the map of the standard book cover,therefore, we can overcome the flaws of traditional Harris corner detection algorithmwhich needs to set a threshold. And then, use the geometric hashing method whichcombined with the sub-regional matching method to achieve an exact feature matchaiming at the detected books and standards cover image. A series of experiments haveverified that the methods in this paper is insensitive to light changes of shooting, booksrotating, and books-scale transform. At the same time, it’s effective to identity thebooks cover which are highly similar.(4) Study the method of recognizing books based on the cover image macthing. By comparing the evaluation method of feature point matching, such as visual method,the comparative evaluation method, and parameter evaluation method, defined thematching rate and established a method to recognize books based on the books imagematching rate. A large number of real books recognition experiments have verified thatthis method has a high recognition accuracy, good stability to recognize the point ofreading books which can be applied to the actual product.