Exploiting BICC Features for Classification of Advertisement Videos Using RIDOR Algorithm

Abstract

The latest trend in the design of electronic embedded systems focuses on technology integration with a domestic utility concept. Automated classification of advertisement videos from the general program videos is emerging as an essential task that is expected to support the television (TV) viewers to have a seamless visual experience without being hampered by commercial advertisement videos (ADD). The demand for a solution is gaining momentum which will enable the viewers to skip the advertisements and move automatically to another channel. This can be achieved just by classifying the extracted frames of videos of advertisements and non-advertisements videos (NADD) which consist of more visual information. In this present work, the descriptive feature is derived through the block intensity comparison technique and applied on 8x8 block size of the frames. The best performing features are identified and selected by decision tree (J48) algorithm and these selected features are used for classification by the Ripple Down Rule Learner (RIDOR) algorithm. The experimental results demonstrate the performance evaluation of RIDOR algorithm, the importance of dimensionality reduction and the comparative study of various classifiers. Therefore, RIDOR achieved the best classification accuracy of 87.12% is reported for further study.