Human Vision Models of Image Quality Evaluation for Jpeg2000 Compression Designed in Matlab

One of the main reasons of image quality evaluation is introduction and employment of the image compression methods and them corresponding formats. The first approach to the image quality evaluation is subjective quality testing (e.g. DSIS Double Stimulus Impairment Scale, DSCQS Double Stimulus Continuous Quality Scale, SCM Stimulus Comparison Method, SSM Single Stimulus Method, SSCQM – Single Stimulus Continuous Quality Evaluation), which is based on many observers that evaluate image quality. These tests are time demanding expensive and have a very strict definition of observational conditions [1]. The second approach is the objective image quality testing (e.g. SNR – Signal to Noise Ratio, MSE – Mean Square Error, MAE Mean Absolute Error) based on mathematical calculations. The objective quality evaluation is easier and faster then the subjective one because observers are not needed [2], but generally these testing have bad correlation (ρ = 0.4 0.7) with objective criteria. The third way of image quality evaluation is usage of a human visual model (HVS) [3, 4]. HVS model combines and uses both the objective and subjective methods. These HVS models can model only parts of human vision that we need (e.g. spatial resolution, temporal motion, color fidelity, color resolution...) [3]. A majority of these models requires a tested image and its corresponding matching reference in order to determine the perceptual difference between them. HVS models can be divided into two groups. The first group comprises one-channel models [2, 3] that can be characterized by computing with the whole image. In the second one there are multi-channel models [2, 3, 4] that simulate the neuron response of the brain cortex. The response is selective to spatial frequencies and orientations. These models decompose the image into the spatial frequency bands and/or orientations. Then, separate thresholds are set for each channel. At the end of the processing the channels are weighted and summed in order to get a number that represents the image quality.