Comparison Between JPEG2000, AVC and HEVC in Compressing Scanned Sheet Music Images

Abstract

This paper evaluates the performance of different image and video coders in compressing scanned sheet music images. For that purpose, the state of the art still image coder JPEG2000 and the video coders AVC and HEVC are used. First, each page of the scanned musical piece is treated as a still image and compressed independently by JPEG2000, AVC-INTRA and HEVC-INTRA. Then, the scanned pages are interpreted as frames of a video sequence and encoded by AVC-INTER or HEVC-INTER. By doing so, interframe prediction may be used as a pattern matcher. Since sheet music has a well behaved structure of symbols, it is expected that interframe prediction will easily find patterns on reference frames that are very similar to those being currently encoded. In other words, present frames use previously encoded frames as a dictionary. The pattern matching algorithm (motion estimation and compensation) generates residual data that can be more efficiently compressed. Results show that HEVC consistently outperforms AVC and JPEG2000. Moreover, the proposed experiments indicate that HEVC-INTER, in average, outperforms HEVC-INTRA when used to compress sheet music images.