ASYMMETRIC FILTER FOR TEXT RECOGNITION IN VIDEO

Stripes are a common sub-structure of text characters, and the scale of the stripes does not vary significantly within a character. In this paper a new form of filter is derived from the Gabor filter which can efficiently estimate the scales of such stripes. The contrast of text in video can then be increased by enhancing the edges of those stripes found to have a suitable scale. The algorithm presented enhances the stripes in three selected scale ranges. Character recognition is then performed on the output of binarizing these enhanced images, and shows improvement over other methods.