Film line scratch detection using neural network

Abstract

Line scratches are one of the most common degradations in old films. To support a demand of high quality of multimedia service, these should be detected and restored automatically. However, although many detection and restoration algorithms have been researched, little have done in automatic scratch detection. This paper presents a texture-based object detection method for scratch detection. We use a multi-layer perceptron (MLP) to automatically make a texture classifier that discriminates between scratch regions and non-scratch ones in various environments. To assess the validity of the proposed method, it has been tested with all kinds of scratches, that is, principal/secondary scratches, alone/not-alone ones, and moving/static ones, and then experimental results show that the proposed approach leads to not only robust but also efficient scratch detection. © Springer-Verlag Berlin Heidelberg 2004.