DEVELOPMENT OF AN IMPROVED KEYFRAME EXTRACTION SCHEME FOR VIDEO SUMMARIZATION BASED ON HISTOGRAM DIFFERENCE AND K-MEANS CLUSTERING

By

Author

Presented To

Department of Engineering

ABSTRACT
The rate of increase in multimedia data necessitated the need for a large number of storage devices. Nonetheless, the stored multimedia data has a lot of redundant video frames. These redundant frames make video browsing and retrieval difficult as well as time-consuming for the user; hence, negatively affecting bandwidth utilization and storage capacity. In order to improve the bandwidth utilization and storage capacity, keyframe extraction algorithms were developed. These algorithms were implemented to extract a unique set of frames and eliminate redundant ones. However, despite the achieved improvement in the keyframe extraction process, there exists a significant number of redundant frames in the summarized video. In order to address this issue, this research presents the development of an improved keyframe extraction scheme for video summarization based on histogram difference and k-means clustering. The developed scheme is suitable for the detection of shot transitions and extraction of keyframes in both low motion and fast-moving videos. The histogram-based approach was utilized to detect shot transitions in the video. Furthermore, the k-means clustering approach was used to efficiently extract a unique set of keyframes. The performance of the developed scheme was evaluated on 4 different videos namely; surveillance footage, movie clip, advert, and sport videos which were all obtained from the popular video-sharing website YouTube. Results were compared with existing schemes of Rodriguez et al.,( 2018) and Sheena and Narayanan (2015) using compression ratio, precision and extraction rates, and f-measure as performance metrics. In terms of the compression ratio, the results showed that the developed scheme outperformed the existing schemes by 24.20% and 35.65%. In terms of precision, it also outperformed the existing schemes by 8.60% and 11.31%. Also, in terms of extraction rate, it outperformed the existing schemes by 0.49% and 7.04%. It also showed an improvement in f-measure by 4.65% and 9.22% when compared with the existing schemes