Research And Application On Key Algorithms Of Chinese Character Recognition

Chinese characters and calligraphy are the main parts of Chinese culture,which constitute the rich and colorful history and culture of China.With the development of science and technology,more and more books or works are digitized to the digital library.When user users appreciate these digital resources,especially for the ancient books,they may meet unfamiliar characters.It will be great if there exists a tool could help users recognize unfamiliar characters and query the meaning of them.This paper presents an algorithm based on GIST feature,SIFT feature and SSC encoding algorithm.Firstly,we collect various dataset of Chinese characters to build a complete character set and create the image library by segmenting.Secondly,the so-called GIST and SIFT features are extracted from the image library to establish feature library.To improve the recognition performance,we filter the SIFT feature points of similar images and reserve the unique feature points of each image to reduce the storage of feature library.Finally,compress the storage of GIST and SIFT descriptors to accommodate mobile platform with Similarity Sensitive Coding(SSC)algorithm.At the stage of recognition,the high dimensional indexing algorithm is applied in feature matching,which improves the efficiency and expansion of the recognition algorithm.The calligraphy recognition is more difficult than Chinese character recognition due to its various forms.This paper proposes a recognition algorithm which combines the deep learning method and traditional recognition method.At the beginning,classify the style of calligraphy image by convolution neural network,then use the traditional classification algorithm to recognize the calligraphy image,in this stage,we user deep feature as the representation of calligraphy image.Through the comparison of experiments,the calligraphy recognition algorithm proposed in this paper is more efficient.Finally,we apply that recognition algorithm in Windows and Android platform,which based on three levels:extracting feature from image,feature matching and system architecture.It shows great performance in recognition and gives smooth experience to users.The Android App developed by us is convenient for users to search unfamiliar Chinese characters by camera.