Fuzzy direction pre-classification based feature extraction using Gabor wavelet transformation

Text image from natural scene is usually low quality because of tilt or low resolution, which causes character recognition degraded. In this paper, a novel method of feature extraction based on fuzzy direction pre-classified is presented that makes use the direction selectivity of Gabor filters. The direction pre-classification of character strokes makes the selection of the direction more accurate. Experimental results show the method adjust to the text deformation.