Video similarity measurement using spectrogram

摘要

A new video similarity measurement method is developed for video copy detection. The ordinal feature is transformed to spectrogram by Short-Time Fourier Transform to represent as a video signature. Similarity between video signatures was measured by using DTW algorithm. The experiments on the CC_WEB_VIDEO dataset show that accuracy of this algorithm is 19.6%, 11.7%, and 16.6% as high as Sliding Window, DTW and STD methods, respectively in both “minor edited” and “extensively edited” video categories.