Video Retrieval by Feature Learning in Key Frames

摘要

We evaluate the application of feature-vector based image retrieval methods to the problem of video retrieval. A vast number of primitive features is calculated for each of the key frames generated by a segmentation process, and we examine the use of three methods for retrieving video segments using the features ― a vector space model, a learning method using the AdaBoost algorithm, and a k-nearest neighbour approach.