Key-frame Extraction and Shot Retrieval Using Nearest Feature Line (NFL)

Abstract

Query by key frame or video example is a convenient and often effective way to search in video database. This paper proposes a new approach to support such searches. The main contribution of the proposed approach is the consideration of both feature extraction and distance computation as a whole process. With a video shot represented by key-frames corresponding to feature points in a feature space, a new metric is defined to measure the distance between a query image and a shot based on the concept of Nearest Feature Line (NFL). We propose to use the "breakpoints" of feature trajectory of a video shot as the key frames and use the lines passing through these points to represent the shot. When combined with the NFL method, it helps to achieve a better performance, as evidenced by experiments.