Commercial detection with hidden markov model

Abstract

Many people who video-record TV programmes find that the advertisements appear regularly quite annoying because they are irrelevant and disrupt the efficiency of browsing and viewing of the recorded programmes. It is also troublesome to manually skip the recorded commercials. This study looked into both feature and learning-based methods so as to for solutions of commercial detection, a process of detecting and logging commercial broadcast from digital video file. In the project, the features such as shot change frequency and frame marked with product information (FMPI), and a learning based algorithm, Hidden Markov Model (HMM), were used for the detection. Ten sample videos were selected for testing of the solutions, which showed the proposed method could detect about 70% to 90% of commercials. In order to see if there were any better learning based algorithm, Support Vector Machine (SVM) was tested in comparison with HMM. However, SVM performed less well in the precision of the detection and HMM remained the better option. In conclusion, HMM generates satisfactory results in commercial detection. Some possible ways are also suggested for further improvement.