MRKFE: Designing cascaded map reduce algorithm for key frame extraction

Key frame extraction is a technique used to compute measure of dissimilarity between successive frames in order to identify unique frames from a collection of video frames and thereby, eliminates redundant frames. It is widely used in image processing, video surveillance, cyber forensic, video browsing, and video indexing and retrieval applications. Though the state-of-the art computer vision algorithms provide efficient solutions for extraction of key frames, but often demands very high computational power as it involves computing very high dimensional image features. On the other hand, with the widespread use of Big Data and IOT applications, the storage, management and accessing of large volume, high velocity and variety of video data is increasing at exponential rates. Therefore, in order to address this problem we propose optimization of key frame extraction algorithms on Apache Hadoop distributed framework. As the process of extracting key-frames involves simultaneous execution of image processing tasks, the Hadoop framework divides and distributes these tasks to multiple nodes of the cluster. The MapReduce programming model has been redefined to perform the tasks of identifying key frames by color histogram difference between successive frames. The proposed optimized Hadoop's MapReduce algorithm has been tested on a data set of 1000 news videos each having approximately a length of 10 minutes. A performance speed up of 7 has been achieved by varying the size of the cluster from 1 node to 10 nodes.