A human visual system based image coder

Over the years, society has changed considerably due to technological changes, and digital images have become part and parcel of our everyday lives. Irrespective of applications (i.e., digital camera) and services (information sharing, e.g., Youtube, archive / storage), there is the need for high image quality with high compression ratios. Hence, considerable efforts have been invested in the area of image compression. The traditional image compression systems take into account of statistical redundancies inherent in the image data. However, the development and adaptation of vision models, which take into account the properties of the human visual system (HVS), into picture coders have since shown promising results.

The objective of the thesis is to propose the implementation of a vision model in two different manners in the JPEG2000 coding system: (a) a Perceptual Colour Distortion Measure (PCDM) for colour images in the encoding stage, and (b) a Perceptual Post Filtering (PPF) algorithm for colour images in the decoding stage. Both implementations are embedded into the JPEG2000 coder. The vision model here exploits the contrast sensitivity, the inter-orientation masking and intra-band masking visual properties of the HVS. Extensive calibration work has been undertaken to fine-tune the 42 model parameters of the PCDM and Just-Noticeable-Difference thresholds of the PPF for colour images. Evaluation with subjective assessments of PCDM based coder has shown perceived quality improvement over the JPEG2000 benchmark with the MSE (mean square error) and CVIS criteria. For the PPF adapted JPEG2000 decoder, performance evaluation has also shown promising results against the JPEG2000 benchmarks. Based on subjective evaluation, when both PCDM and PPF are used in the JPEG2000 coding system, the overall perceived image quality is superior to the stand-alone JPEG2000 with the PCDM.