Understanding shape context by analysis of Farey ranks

摘要

A novel method for shape context realization has been presented in this paper and specific application of it in hand-drawn character recognition has been shown. In preprocessing stage, the endpoints of the digital straight line segments defining the object boundary are extracted. The four vertices of the minimum area rectangle bounding these extracted points are used as the reference points for shape context analysis. Slopes of the extracted endpoints with respect to the reference points are represented as the ranks of corresponding fractions in Farey sequence. From the distribution of the Farey fraction ranks, we obtain a histogram giving the shape context of the object. Finally, we have presented a modified Chi-square histogram match finding method. Experimental results are quite encouraging in terms of accuracy and robustness.