Research On Chinese Character Recognition In Natural Scene Images

It is more and more convenient for people to obtain images in natural scenes with the popularity of electronic intelligent devices,and the research and recognition of Chinese characters in natural scene images has gradually become a hot issue.Aiming at the problems of low recognition efficiency,low accuracy and poor fitting in some current methods,this paper analyzed and studied the detection,location and recognition of Chinese characters.In the detection and location of Chinese characters,we adopted the edge-enhanced maximum stable extreme region method,used the HOG operator to extract the gradient information of Chinese characters in natural scene images,enhanced the contrast of Chinese characters regions,and combined color clustering and heuristic rules to further eliminate the background information in the images.And then,we used the method of stroke width transform to fine cut the candidate region of Chinese characters,adjusted the angle range of gradient direction in the original method,and combined the closed operations in morphology to eliminate the tiny holes inside Chinese characters.Finally,we calculated the variance of stroke width and coefficient of variation to extract and merge the connected region of Chinese characters.The experimental results showed that the method in this paper exceled in accuracy and recall,and the comprehensive F value was significantly higher than other algorithms.In the recognition of Chinese characters,we introduced the subtractive clustering algorithm to determine the K-means algorithm's initial clustering center and its number,and then used K-means algorithm to determine the radial basis function center of RBF neural network,after that,a Chinese character classifier was designed with the improved RBF neural network.Secondly,a 12-layer deep convolutional neural network model was designed to recognize Chinese characters blocks.Finally,we adjusted and modified the previous results comprehensively in the post-processing of Chinese characters recognition.The experimental results of the two groups showed that the Chinese characters classifier in this paper had a higher accuracy rate,and the deep convolutional neural network model had a better fit for Chinese characters.