Study on video text processing

Abstract

This paper adopts combination of DWT and neural network methods to locate the text region. After analyzing the characteristic of video text, the feature of kurtosis is first used in the field of text extraction, which improves the locating accuracy considerably. The OCR accuracy is also improved by using the Shannon interpolation method and Niblack adaptive thresholding method to enhance the video text extracted. Experiments show the methods mentioned above improve the locating accuracy to about 90.4% and the OCR accuracy to about 85.1%.