Research On Book Spine Recognition Algorkhm Based On Computer Vision

At present,with the increase of people's library,library management is becoming more and more difficult.The shelf manual management of books is a tedious task for humans,if this work can be done by robots,the human work will be greatly reduce.But the gap between human and computer in the image positioning and search is very large.Book management is usually based on RFID technology,This paper studies the existing algorithms for segmentation and recognition of the spine image,propose a book management algorithm based on computer vision.This algorithm includes the spine contour extraction,image matching,text location.Analysis of the spine contour extraction algorithm based on book spine features.This paper proposes a optimized book spine segmentation algorithms which overcomes the defects of existing book spine segmentation algorithms that can't segment the book spines accurately when there is gap between the book spines or the book spines are disordered placed.The proposed algorithms can achieve better segmentation results by capturing the candidate book sine border based on gradient of edge growing algorithms,then combining the proposed filtering algorithms based on maximum independent sets of candidate book spines.The test results show that the proposed algorithms not only has better segmentation results when comparing with the existing segmentation algorithms,but also it can fix the issue that the existing segmentation algorithms can't segment the book spines accurately when the book spines are disordered placed.This paper analyze the text location algorithm based on the scene.According to the characteristics of the spine image,this paper proposes a text location algorithm based on MSER feature.text filtering algorithm base on the characteristics of the spine image text information,which can accurately recognize the text information.Analysis of the spine recognition algorithm based on image.However,since most of the existed feature matching are not so powerful and efficient when the image after affine transformation.The underlying reason is that a suitable measure is unavailable currently in literature to adequately evaluate the similarity of two images.In contrast to the traditional evaluation methods,in this paper,the registration evaluation problem is converted into a classification problem through machine learning.Experiments show,the validity and reliability of our proposed method.