VIDEO TEXT EXTRACTION USING DISCRETE WAVELET TRANSFORM

Abstract

Text extraction from video content is considered as a key part in understanding, structuring, mining, indexing, automatic annotating and retrieval of useful information. However, automatic video text extraction is extremely challenging due to variations of text, low image contrast, and complex background. In this paper, we have proposed a discrete wavelet-based method for localizing and extracting caption or superimposed texts from videos. First, the vertical, horizontal, and diagonal edges are detected by using Haar wavelet transform. Then, region of interest (ROI) is extracted by logical AND operation of edge images. After that, adaptive run length smearing algorithm (ARSLA) along with projection profile analysis is exploited for noise refinement. Finally, text regions are localized and extracted using features of connected components. Experimental results confirm the satisfactory localization and extraction of captions and subtitles in video sequences using the designed approach.