A Study on various Key Frame Detection Techniques in order to Develop an Efficient Method

Abstract:

A huge amount of data is generated in the form of text, images, and blogs globally. Video processing has gained immense significance due to advancements in network architectures, high availability of storage, and easy access to digital cameras. However, due to the remarkable development of multimedia and Internet technology, high resolution digital video has gradually replaced monotonous text information, which has become one of the main ways for people to propagate information. Each video consists of numerous frames that contain essential information. It is crucial to extract these frames to process the video. Retrieving key-frames from all the frames, which contain distinguished features of the video helps us in developing advanced technologies that will help in numerous video analytics applications like object and anomaly detection. This paper compares traditional key-frame extraction methods – Clustering, Motion Analysis, and Shot-Boundary based, as well as deep learning key-frame extraction methods that have been currently implemented. The aforementioned methods have been implemented and a detailed comparative analysis has been presented along with their pros and cons.