A Robust Algorithm for Arabic Video Text Detection

Abstract

In this paper, we propose an efficient Arabic text detection method based on the Laplacian operator in the frequency domain. The zero crossing value is computed for each pixel in the Laplacian-filtered image to found edges in four directions. K-means is then used to classify all the pixels of the filtered image into two clusters: text and non-text. For each candidate text region, the corresponding region in the canny edge map of the input image undergoes projection profile analysis to determine the boundary of the text blocks. Finally, we employ empirical rules to eliminate false positives based on geometrical properties. Experimental results show that the proposed algorithm is able to detect texts of different fonts, contrasts and backgrounds. Moreover, it outperforms four existing algorithms in terms of detection and false positive rates.