Detection and Removal of Scratches in Images

摘要

In this paper a detection/restoration method to detect and remove line scratches in still images, regardless of their orientation, colour, and shape is presented. In this work the two properties of scratches are considered namely: scratches have high contrast compared with its neighbours and they usually occur vertically and it is more than half of the image in length are taken into account. Firstly a simple thresholding technique is applied for detecting the candidate scratch pixels. The pixels detected in the first step are then used as a mask for removing the scratch during the restoration step. The detected scratch is inpainted using our higher order non-adaptive interpolation approach based on a 13×13 neighbourhood. The experimental result shows that the proposed method works well for both simple and complex scratches that are present in uniform or less complex backgrounds. When experimented with complex backgrounds the interpolation artefact namely blurring becomes pronounced. The proposed work can be effectively used to automatically detect and remove the scratch from uniform and less complex static images without user intervention in any stage of the process.