Research And Design Of Document Image Stitching System Based On AM5728

With the continuous development of image processing technology,image stitching has become a hot spot in the current research,which is widely used in various fields,such as remote image,resolution enhancement,global monitoring,and medical imaging.The digital documentation is widely used because of its easy to management and storage.During the document digitization procedure,a single image is insufficient to gain a whole image clearly.Therefore,several images are acquired for stitching to obtain a complete and clear image.In this thesis,the document image preprocessing is carried out on the document image captured by the camera at first.Then,the document image mosaic algorithm is proposed.Finally,a set of document image mosaic software is designed and developed based on the TI's high-performance multi-core video processor AM5728.The main work of the thesis is summarized as follows:(1)The background and the significance of this subject are illustrated in the thesis.Besides,the existing image registration and image blending algorithms are both deeply explored.Afterwards according to the document image mosaic system,the main research contents and key points are concluded.(2)The process of image sensor selection is firstly introduced combined with document image features.Subsequently,the automatic exposure method is described by combining the stepwise approximation method and the look-up table method.However,some regional shadow areas are still remained after automatic exposure,a shadow compensation approach is proposed by using the area where the exposure of image is normal.For the prevalence of salt and pepper noise in document images,a self-adaptive window denoising method is used to remove the salt and pepper noise effectively while preserving the document image information.In addition,in order to obtain a better display result,image correction is performed by the gamma curve.(3)In the document image stitching,the document image feature points are over extracted and the matching error rate is quite high.In this thesis,the feature vector of each connected domain is obtained by using the angular transformation(ART)after concatenating the domain,which effectively reduces the false matching rate and improves the image registration accuracy.The document image foreground and background differences easily cause ghosting problems.Hence,a document image blending method is proposed in this thesis based on the optimal splicing.In this method,overlapping regions are acquired by image registration at first.Then the projection method is used to detect the smallest difference area in the binarized document image horizontally and vertically.Finally,the best splicing seam is found for document image blending.(4)In order to make the algorithm more suitable for practical application,the algorithm is transplanted to the TI's high-performance multi-core video processor AM5728 development platform.The overall scheme of the hardware and software and the framework of the document image mosaic algorithm are introducted.OpenCL technology is used to parallelize the design of the key algorithm to improve the operating efficiency of the algorithm.Experiments show that the parallelization of the application effectively improves the operating efficiency of the algorithm.