Extraction , Enhancement and OCR

In this paper we address the problem of text extraction, enhancement and recognition in digital video. Compared with optical character recognition (OCR) from document images, text extraction and recognition in digital video presents several new challenges. First, the text in video is often embedded in complex backgrounds, making text extraction and separation diicult. Second, image data contained in video frames is often digitized and/or subsampled at a much lower resolution than is typical for document images. As a result, most commercial OCR software can not recognize text extracted from video. We have implemented a hybrid wavelet/neural network segmenter to extract text regions and use a two stage enhancement scheme prior to recognition. First, we use Shannon interpolation to raise the image resolution, and second we postprocess the block with normal/inverse text classiication and adaptive thresholding. Experimental results show that our text extraction scheme can extract both scene text and graphical text robustly and reasonable OCR results are achieved after enhancement.