Lossy compression of natural HDR content based on multi-component Transform optimization

Research output: Contribution to journalLiterature reviewResearchpeer-review

Abstract

Linear multi-component transforms (MCTs) are commonly employed for enhancing the coding performance for the compression of natural color images. Popular MCTs such as the RGB to Y'CbCr transform are not optimized specifically for any given input image. Data-dependent transforms such as the Karhunen-Loève Transform (KLT) or the Optimal Spectral Transform (OST) optimize some analytical criteria (e.g., the inter-component correlation or mutual information), but do not consider all aspects of the coding system applied to the transformed components. Recently, a framework that produces optimized MCTs dependent on the input image and the subsequent coding system was proposed for 8-bit pathology whole-slide images. This work extends this framework to higher bitdepths and investigate its performance for different types of high-dynamic range (HDR) contents. Experimental results indicate that the optimized MCTs yield average PSNR results 0.17%, 0.47% and 0.63% higher than those of the KLT for raw mosaic images, reconstructed HDR radiance scenes and color graded HDR contents, respectively.

Original languageAmerican English
Pages (from-to)23-28
Number of pages6
Journal2016 Digital Media Industry and Academic Forum, DMIAF 2016 - Proceedings
DOIs
Publication statusPublished - 22 Sep 2016

Keywords

  • HDR
  • Image Compression
  • Multi-Component Transforms

Fingerprint Dive into the research topics of 'Lossy compression of natural HDR content based on multi-component Transform optimization'. Together they form a unique fingerprint.

Cite this