Video Scene Change Detection Using Convolution Neural Network
Sung-Jun Han, Jongweon Kim · 2017
There are various contents such as music, image, and image on digital. There are many kinds of dual image contents such as movies, drama, entertainment, and anime. In the case of image contents, a frame or a shot unit, which is one image unit, a scene where consecutive frames are gathered, and a plurality of scenes are connected to form a single image. In this paper, we propose an algorithm to distinguish scenes from image contents. We distinguished the scene using a convolution neural network for video scene discrimination. The video was classified by frame, and the data of the frame portion in which the scene such as the frame data of the portion where the scene changes was held was labeled and trained. By using the trained network, it is possible to extract images in frame units from movie contents such as movies and drama, and to identify the scene changing parts and to divide the images into scene units. It is possible to distinguish scenes with higher accuracy even in areas that change very quickly, such as actions, compared to conventional methods.