Abstract: Feature extraction and classification method was presented based on statistics template for classifying halftone images produced by various error diffusion methods．Statistics template was described as the descriptor of texture feature of halftone image according to the definition of pixel pairs,and a feature extraction method was presented based on image patches．The ideas of class feature matrices was proposed acting as the descriptor of category and then the optimization problem was formulized by establishing error object function and utilizing gradient descent method to seek the optimal class feature matrices．The characteristics of class feature matrices were discussed by experiments．In experiments,the performance comparisons of our method with two similar methods were conducted．The influences of parameter on classification performance were also discussed and time complexity of feature extraction algorithm was analysed．Experimental results demonstrate that the proposed method is effective.

Key words: Halftone image,Error diffusion,Statistics template,Classification