Removing Labeled Text in an Image Based on an Extended Filter in Gabor Space

In this paper, a blindness method using an extended filter in Gabor space for removing the text labeled by users in an image is proposed. In order to eliminate the text region, firstly, the features of an image are calculated in Gabor space, and then Gabor filter with different directions and frequencies are applied to the images. Since Gabor filter allows us to obtain channels in different scales and angles, and the relationship of different channels is a core factor which used in calculating coefficients to find possible labeled area because that in contrast to the original image, these channels in Gabor space can present the characteristics of text regions well, so the presented algorithm can remove the text area effectively. To demonstrate the validity and efficiency of the method, the experiments on the real-world images from a large number of webpage demonstrate are presented in the last.