Study On Segmentation And Recognition Technology Of The Low Quality Pressed Characters

Study on optical character recognition technology is more and more mature at home and abroad, the segmentation and recognition of protuberant character using optical character processing, but in practice it is still need to improve. In this paper, we analysis of the characteristics of the pressed characters and view the research status at home and abroad, use the image guided filtering algorithm to preprocess the character. The algorithm originated in the linear model, by the guiding role of guiding image to input image, it makes the output image improved in detail. Compared with the traditional mean filter algorithm, the image guided filtering algorithm can make better image detail enhancement, it will not weaken the useful edge information at the same time. Through the comparison of the output image and histogram, the effectiveness of the algorithm is very obvious.In the segmentation, the adhesion protuberant character segmentation using the drop-fall algorithm, the principle is to simulate the process of water droplets falling, the most important thing is to determine the starting point and cutting path. In the selection of the starting point, method is scanning the black and white pixels, and the cutting path is the traditional cutting path of the drop-fall algorithm, it can segment many adhesion protuberant characters. But there are some small problems, small grooves caused by local small burrs generate pseudo cut point, the algorithm designed uses the neighborhood scan algorithm to determine the bifurcation point and the convergence point, changing the cutting path from dripping close contour to dripping directly on the corner, and after improvement, the results of segmentation are more accurately.Finally using the HOG algorithm and SVM to do the feature extraction and recognition of characters, this paper extracts the edge information of character contours by the HOG feature extraction algorithm which used for face recognition currently, and it shows the more superior effect than Sobel algorithm. Through the comparison of different parameters, the paper selects some of more appropriate parameter value to have better effect in the experiment. From the overall significance of the research, this article can contribute to the overall study of methodological improvements of lower quality platen character, it has further improved to the protuberant character segmentation and recognition, and it makes contributions to subsequent research and improvement.