Image-based Table Recognition System Of Printed Documents

The recognition of printed document images,such as books and reports,is an important branch of pattern recognition.The technology of character recognition has been mature,and there are many commercial products in the market.The identification technology for the printed form image has yet to be further improved.The existing methods require high degree of regularity and poor robustnes s in the recognition,and most recognition systems are based on the PC.With the development of the mobile Internet,the PC-side system has been unable to meet the demand.Based on the analysis of related technology and product status at home and abroad,this paper studies the preprocessing methods of table images,the recognition of closed printed forms,the recognition of Chinese characters in printed forms,and the design of Android-based form recognition software.A highly robust form recognition system was implemented and the development of the mobile system was completed.The main work of the thesis is as follows:The preprocessing algorithm of table image is studied.The document image is binarized by the optimized binarization algorithm.The document image is tilted and perspective corrected according to the different tilt conditions of the table.The morphology based image is applied.The algorithm identifies and locates the table area and refines the obtained table frame.A cell location algorithm based on contour exclusion filtering is proposed.The algorithm can accurately find the specific location of each subdivided c ell and send the partial ROI to the recognizer for identification,which improves the efficiency of cell identification and positioning.The printed Chinese character recognition algorithm bas ed on deep neural network is designed to identify the whole table cells and return the results,which avoids the row and column character segmentation operation and reduces the interference introduced by segmentation.The experimental results show that the designed deep neural network is based on the neural network.The printed Chinese character recognition algorithm can achieve 96% accuracy.The table recognition algorithm is realized by C++.The JNI technology is used to integrate its call into the Android system,which improves the execution efficiency and portability of the algorithm,designs a friendly app interaction interface,and completes the software development based on Android,achieve a mobile-based form recognition system.Based on the above algorithms,the paper realizes the recognition system of printed form based on Android system.After experimental verification,the designed form recognition system acquires the form image in the natural scene,which can identify high-quality forms very well,and the recognition rate of low-quality forms such as distortion and blur can reach 81%,which is higher than t he commercial form recognition software.