Using JPEG and JPEG2000 Compressions for Fast Image Quality Metrics Based on Free Energy Theory.
Ke Gu,Guangtao Zhai,Min Liu,Xiaokang Yang,Wenjun Zhang
DOI: https://doi.org/10.1109/bmsb.2014.6873472
2014-01-01
Abstract:In this paper, we propose a fast, effective and practical algorithm for image quality assessment (IQA). Recently, a new free energy theory was revealed in the field of brain science, which illustrates that the human visual system (HVS) always strives to comprehend the input visual signal by reducing the undetermined portions. Inspired by this, our previous work recently designed a valid reduced-reference (RR) free energy based distortion metric (FEDM) using linear autoregressive model. Despite of fairly well performance, the FEDM is yet difficult to work in real-time applications owing to its weak portability and considerable computational load. Using an alternative way, this paper approaches the free energy related mechanism in human brains with JPEG and JPEG2000 compressions. The proposed metric can work quickly and practically in that it is realized with highly developed and widely employed image compression methods. Results of experiments on the most popular publicly-available LIVE database demonstrate the effectiveness of the proposed RR algorithm over classical full-reference PSNR and SSIM as well as stateof- the-art RR IQA metrics.