Tree structured vector quantization with dynamic path search

Abstract

A new branch of tree-structured vector quantization is proposed to encode images. We call it the dynamic path tree structured vector quantization (DPTSVQ). Multipath TSVQ uses a fixed number of paths to search the closest codeword; however, there is still plenty of room for improvement. To do way with the lack of flexibility fixed number of search paths, we propose DPTSVQ to take the place of multipath TSVQ. With DPTSVQ, we try to improve multipath TSVQ and make the number of search paths become variable. In this paper, we define a critical function to judge whether the number of search paths is growable for DPTSVQ. Our experimental results show that DPTSVQ is always faster than multipath TSVQ with the image quality kept the same. DPTSVQ can reduce 50% of the encoding time, in general, from what is spent by multipath TSVQ under the same image quality requirement. If such high image quality as that of full search is required, DPTSVQ remains more timesaving than multipath TSVQ all the same.