Text detection and removal from image using inpainting with smoothing

In this paper we have implemented a two stage frame work to remove unwanted text from images: first stage is to detect text from image and second stage is to remove that text using method of inpainting. To detect text, text localization and extraction is carried out followed by method of inpainting to fill the holes generated in image using surrounding region. To bring more efficiency in text detection smoothing works better. With the use of feature extraction, stroke filtering and centroid processing text detection is performed. Color histogram processing is carried out to carry clearer filtering of text components from other image components. In last stage, the text holes generated are filled with appropriate information present in same image using nearest matching neighborhood inpainting. To generate visually plausible region filling results, smoothing is carried out on the selected filled patches. We tested the implementation using different images, and compared the results of smoothing with results of implementation without smoothing. Experimental results show improved PSNR due to smoothing.