Research On Book Association Detection Based On Signature Marks

Book association detection based on signature marks uses image processing to complete detection task based on the information of signature marks in the spine images captured by a CCD camera.This paper has important engineering and theoretical significance for improving the quality of bookbinding,the production efficiency and the industrial automation.This paper divides the research on book association detection based on signature marks into three fundamental links: signature mark detection,sewing region segmentation and book association matching.(1)Signature mark detection.Inspired by the non-classical receptive field mechanism,a signature mark detection method based on the three-Gaussian model is proposed.This model is essentially a band-pass filter for background suppression,target enhancement and noise filtering.In the signature mark detection method based on the three-Gaussian model,the parameter selection method is constructed for the three-Gaussian model according to the prior knowledge of the distribution characteristics and statistical characteristics of the targets,the background and noises;Then,a compensation measure factor is introduced to enhance the edges and smooth dark background regions.Experiments have shown that our method is effective.(2)Sewing region segmentation.An edge-based method of sewing region segmentation is proposed,which includes the rough segmentation method based on the K-means clustering algorithm and the accurate localization method based on edge fairness.In the step of rough segmentation of sewing regions,first extract the vertical edge features of a spine image to generate a horizontal edge image and a feature point set;Then,calculate the maximum points of the horizontal edge histogram after the secondary filtering of the S-G algorithm to determine the number of sewing regions.Finally,the K-means algorithm is used to cluster the feature point set.In the step of accurate localization of sewing regions,first select the initial edge of the optimal detection region according to the shape description metric based on convexity;Then,select the edge points with large curvature fluctuations on the initial edge,adjust the position of the selected edge point according to its neighboring points to minimize the curvature fluctuations of this selected edge point and its neighboring points,and stop the adjustment process when the edge smoothness measure is constant;Finally,the accurate edge of the optimal detection region is translated horizontally to obtain the accurate edges of other sewing regions,so as to obtain the final accurate sewing regions.Qualitative analysis and quantitative evaluation of the segmentation results of sewing regions have shown the effectiveness of the proposed method.(3)Book association matching.On the basis of the signature mark detection results and the sewing region segmentation results,the signature mark restoration,signature mark recognition and shape matching cost calculation are performed.First,restore the signature marks according to whether a signature mark is intersected with its nearest sewing region;Then,construct the column gray curve to calculate the thickness of a signature,and extract the largest connected domain in each signature as a true signature mark;Finally,the shape context algorithm is used to calculate the shape matching cost of the signature marks in the template spine image and the spine image to be detected,and the correctness of the book association is judged according to the matching cost.The experimental results show that the book association detection method designed in this paper can meet the accuracy requirements of the industry.