Multi-feature Multi-pass Dissolve Detection

This paper introduces a new dissolve detection approach focusing on the detection of difficult and slow dissolves, whose reliable detection is still considered difficult up to today. The improvement presented contributes to the existing work based on the use the variance of pixel intensities to detect fades and dissolves. We propose a multi pass approximation of the linear behavior of the mean signal and the semi parabolic behavior of the variance signal for slow dissolves while introducing a detection based on the introduced exp feature for especially fast dissolves and then combining their detection results. Focusing on a multi feature and multi pass approach we aim to improve overall recall and precision rates for dissolves. This is achieved by reducing the number of false positives to a acceptable rate by combing two precise specialized features and detection methods. Experimental results show that the proposed approach provides a 85.96% precision and a 84.01% recall for the selected real TV videos.