IMPROVED-SDROM filtering for scratches removal from images

摘要

This article is dedicated to the detection and correction of scratches found in old movies. The method we propose is based on the SDROM method (Signal Dependent Rank Ordered Mean) which corrects only the pixels of the detected scratches. A statistical study of the noised images by scratches shows that the amplitude difference of the neighboring pixels in an image outside the scratch is small (less than 10 Gy levels). We find that the scratch can be characterized by a high difference of amplitude of its edges pixels, we propose an approach called IMPROVED SDROM, is constituted by three stages. First for the detection of the scratches we use two neighboring sliding windows (3 × 3 pixels) sweeping the entire image. We show that Δ = m_2-m_1 (with m_1 and m_2 the averages of the two windows), is a relevant parameter for the detection of pixels that can belong to a stripe. The average of each window is calculated after a pre-treatment. Then we locate the stripes and their widths and finally we make the correction by a simple interpolation. Unlike the SDROM method, our approach allows to locate scratches of any width, with simplicity of treatment certainly allows a gain in processing time compared to other methods that will be mentioned in what follows. A study of a set of examples of scratches obtained by simulation and also on real scratches illustrates the validity of our approach.