Meter Character Recognition Based On Neural Networks

Water meter automatic recognition system is a new application of artificial intelligence technology in life. One of its vital components is the water meter image character recognition. According to the trait of water meter character, the article divides it into three steps to implements the water meter image character automatic recognition by using the technology of image processing and neural network including. They are water meter image preprocessing, feature extraction, and character recognition using BP neural network.Image pre-processing technique includes image enhancement, binarization and skew correction. In this article, histogram equalization method and intermediate value filter technique are applied for image enhancement. In order to overcome the noise caused by distributing uneven brightness, a binarization algorithm based on Canny's theory is adopted to treat with water meter image.It also presents the thinning method used by this article.Firstly the article summarizes the general methods with regards to character feature extraction. Then it analyzes the trait of water meter character. Full character and deformity character are two kinds of water meter character. In order to extract feature of full character, it uses a method called coarse grid combined counting times the line across character. To deal with the deformity character, it divides the character image into two sub-images. And get a ratio according to the heights of two sub-images. Then let each sub-image extendits size to the size of one full character.Those three sets of value is used as deformity character feature.It uses BP neural network as the classifier to recognize the characters here .After summarizing the notes of BP design, it designs a 3 layers BP nn for this recognition system. Then a great deal of experiments and comparisons have been done, which show the feasibility of the proposed recognizing methods. Comparisons between the classifier proposed here with other classical classifiers proves that BP neural network has strong ability of classify.Summarize and vistas are shown at the end of this article.