Study Of Recognitizing Printed Chinese Characters On Packaging Boxes

It's very important for both producers and users that the explanative characters on the packaging boxes are correct. In order to monitor the characters on the packaging boxes which are flowing on the product line, study is carried out on how to recognize the Chinese characters printed on the packaging boxes.Usually, the character images of the packaging boxes are of low quality because of printing or imaging. Therefore, traditional binarization-based feature extraction and recognition algorithms can not be directly applied. A two-level feature extraction and classification method is used to recognize Chinese characters on packaging boxes.Gaussian filter and thresholding are used in normalizing low quality Chinese characters, and the normalization result is robust enough to the noises.When preliminary classification, Gabor filters are used to extract robust features. A Chinese character can be decomposed by Gabor filters into four sub images with different oritentational strokes, which are corresponding to horizontal, left-diagonal, vertical and right-diagonal. It is easier to design Gabor filters for decomposing strokes than for stroke extraction. Preliminary classification feature is further generated by calculating stroke cross count and DCT transformation. Experiments show that K-means clustering algorithm can reach a preliminary classification rate of 98%.Hidden markov model (HMM) is used to model the Gabor filtering outputs of Chinese characters when fine classification. It is feasible to model two dimensional Chinese characters by one dimensional HMM after Gabor decomposition. Four Gabor filtering outputs are corresponding to four distinct states. An intuitionistic quantization method is proposed for generating observation sequences. The physical meanings of the states and observations are clear and thus ease the initializing of the model parameters.Experimental results show that combining Gabor filter and HMM can well recognize low-quality Chinese characters on the packaging boxes. The final recognition rate reaches 95%.