An exploration of wavelet transform and level set method for text detection in images and video frames

Abstract

In texture-based text detection method, text regions are detected by obtaining textural properties of an image. In order to obtain textural properties of an input image, the proposed system performs single-level 2D DWT. The resultant detail coefficients are averaged to get a better texture properties and to localize for further processing. Then, 2D DWT is explored with a level set method to address the problem of text detection especially curving portions of text present in images and video frames. Thus, the proposed system implements the level set method to detect the true text regions effectively based on contours in images and video frames. Experimental results prove that the proposed level set based method is competitive when compared with other existing methods in reducing false positive rate and mis detection rate. Hence, the proposed system is encouraging and useful to carry out further research on text extraction in images and video.