Abstract:

A classification method of halftone image over supervised manifolds learning were proposed for classification of halftone image according to the characteristic of halftone pattern. The autocorrelation coefficient of pixel on three direction and XNOR operation were used to act as the descriptor of texture feature. A feature extraction based on image patches were presented for reducing modeling time and improving feature efficiency. To enhance the discriminating ability of samples, the linear dimension reduction was conducted in high-dimensional feature space via supervised learning and creating model of noise sample pairs. In experiments, the efficiency problem of feature modeling was analyzed and the  performance comparisons were conducted between the proposed method and five similar methods. The influences of the two parameters on classification performance were also discussed. Experiment results showed that our methods could  get good classification performance if parameter K=32 and L=8 and was  superior to other five classification methods.

Key words: inverse halftoning, feature modeling, halftone image, dimension reduction, classification