Fast Template Matching Using Product Quantization and Its Application to Motion Compensated Prediction

Abstract

Recently, a new motion compensated prediction method that performs template matching between reconstructed images is explored for efficient video coding. Though the method can be carried out at the decoder side and thus requires no motion information to be transmitted, its application to a real-time system is practically limited due to complexity reasons. To cope with this problem, a fast template matching algorithm using a product quantization framework is proposed in this paper.