Detecting abrupt scene change using neural network

Abstract

A real-time algorithm is proposed for the detection of abrupt scene changes, which makes use of a dual (one big and one small) window and single-side checking to avoid the false detection and miss detection caused by the violent motion of camera and/or large objects. In addition, a multi-layer perceptron is used to solve the problem of parameter determination in the proposed algorithm. The performance of our algorithm has been experimentally compared with that of some typical methods by using real video sequences. The recall rate is greatly improved while keeping high precision rate.