Film line scratch detection using neural network and morphological filter

This paper presents a scratch detection method that automatically detects all kinds of scratches from each frame in old films. Generally, the scratch in old films has lower or higher brightness than neighboring pixels in its vicinity and it usually appears as a vertically long thin line. The proposed method is designed from these characteristics of a scratch, thus it consists of two major modules: a neural network-based texture classifier and a morphology-based shape filter with multiple structuring elements. First, the NN-based texture classifier divides the input image into scratch regions and non-scratch regions using the texture property of the scratch. Secondly, the morphology-based shape filter confirms the classified scratch region with structuring elements which is designed based on the shape characteristics of scratches. To assess the validity of the proposed method, the experiments have been performed on several old films and an animation, then the results confirms that the proposed method can detect all kinds of scratches and have the potential to be applied to the commercial systems.