A Novel Dynamic Rate Based System for Matching and Retrieval in Images and Video Sequences

In this paper we describe a system for improved object matching and retrieval in real world video sequences. We propose a dynamic frame rate model for selecting the best frames from a video based on the frame quality. A novel blur metric based approach is used for frame selection. A method for calculating this no-reference blur parameter by selecting only the features contributing to the real blur and discarding unwanted ones is proposed. Precision and Re-call figures are provided for trademark matching in sports videos. Results show that the proposed technique is an efficient and simple way of improving object matching and retrieval in video sequences.