Crease Detection And Distortion Removal Of Scanned Document Images

Compared with the traditional paper documents,the electronic documents are much easier for the storage and transmission.The most commonly approach to digitalize the paper documents is to obtain their images by scanning the documents.Different from the flatbed scanner,the high-speed photographic scanners(HPS)are able to obtain the image in the contactless manner from the precious documents such as ancient manuscripts and books.Since the HPS does not press the document onto the flat plane,the obtained image often suffers from distortions like crease and shading.These artifacts reduce the aesthetic quality and bring difficulties for subsequent analysis and recognition procedures.This thesis presents researches on crease detection and distortion removal for large size document images.The proposed method is able to locate the crease pixels accurately and correct the corresponding intensity and geometrical distortions effectively.The proposed method benefit the subsequent document analysis and recognition by improving the quality of the scanned image.First,this thesis introduces the existing research of related problems including document border detection,shading extraction and document dewarping.Then,based on the existing document border detection and shading extraction,a path searching based crease detection method is proposed for the scanned image of the large scale document.And the the intensity and geometric distortions caused by the crease are removed according to the shading distribution on both sides of the crease.The main contents of our work are as follows:1.For the crease detection,a path searching based method is proposed to locate the crease in the large size document image.The proposed method first coarsely locate crease region by using the intensity feature.Then,a convex hull based level set algorithm is used to extract the shading image of the crease region,on which candidate crease paths are obtained by using a series of filtering,bianarization and morphological operations.Finally,a Dijkstra path searching algorithm is adopted to obtain the accurate path of the crease.Experimental results on the scanned images of newspapers shows the effectiveness and robustness of the proposed method for both straight and crooked creases.2.In terms of crease correction,the illumination and geometric distortion of the image are corrected.After the shading is extracted,the illumination of the image is corrected by intrinsic image decomposition.For the geometric distortion near the crease,a distortion removal method based on shading pixel distribution is proposed.This method extracts the block of the document near the crease according to the location of the crease.The difference of shading pixels between the image near the crease and the normal area is obtained.The scale factor caused by the crease is estimated.The image near the crease is resampled with the corresponding scale factor to remove the geometric distortion.The experimental results show that the proposed method can remove the distortion in the crease region effectively and restore high quality scanned images.