Stone Image Segmentation And Extraction Of The Chinese Characters

Image segmentation is a very important part in digital image processing, it has awonderful use in many regions, and there exist varieties of segmentation methods. Ourresearch is aiming at the stone images, and we have done plenty of experiments, thenwe choose some given methods, and refine some algorithms in our research.According to the characters of the stone images, we classify them. Firstly, becauseof the nature and the human factors, the stone images which we obtain may have manynoises, it’s necessary that we must prior use some filter methods to eliminate thosenoises. Our research chooses some common filters, and then, according to the images’complexities, we have used several image segmentation methods.For the stone images whose distinctions between the Chinese characters and thebackground, we directly use a threshold method called OSTU method, and there is ameaningful usage using OSTU method to segment those gray histograms whosesplit-blip models are unobvious. And then, we choose a method related to the Chinesecharacters called based on Run-Length segmentation method to process those remainednoises coming from the OSTU method and we add a region processing method.Experiments show that based on Run-Length segmentation method is better thanthreshold methods in processing some complex stone images.Some stone images have abundant textures, to these images, we use some texturemethods and then import an advanced texture method called Local Spectral Histogramsfor the first time, the results aren’t very ideal. Through lots of experiments, we find thatthe original algorithm doesn’t contain the direction elements, but the character strokeshave obvious ones. According to this situation, we present a refined algorithm to solvethe direction problem. We introduce four direction elements in the refined method, andcalculate four rectangles in these four directions. Experiments show that the refinedalgorithm is better than the original one in processing stone images.Those previous methods can only process the gray images, and the gray imagesmay lose lots of information compared to the color images. Considering this situation,we particularly process some color images using the RGB Color Space segmentationmethod. And experiments show that it’s useful for some color stone images.We establish a human-computer interaction interface to quickly and efficientlyprocess these complex and variable stone images. This interface contains the mainmethods that we have used and still contains some manual handling methods.