KS-SIFT: A Keyframe Extraction Method Based on Local Features

In this work we propose a new key frame extraction method based on SIFT local features. We extracted feature vectors from a carefully selected group of frames from a video shot, analyzing those vectors to eliminate near duplicate key frames, helping to keep a compact set. Moreover, as the key frame extraction is based on local features, it keeps frames latent semantics and, therefore, helps to keep shot representativeness. We evaluated our method in the scene segmentation context, with videos from movies domain, developing a comparative study with three state of the art approaches based on local features. The results show that our method overcomes those approaches.