Image Colour-Quality Modelling for Mobile LCDs
Youn Jin Kim,Jangjin Yoo,M. Ronnier Luo,Peter A. Rhodes,Vien Cheung,Stephen Westland,Wonhee Choe,Seong-Deok Lee,Changyeong Kim
DOI: https://doi.org/10.2352/cic.2006.14.1.art00029
2006-01-01
Abstract:This paper describes the development of an image colourquality model based on individual physical image statistical measures for mobile liquid crystal displays (LCD). Five natural images were colour-rendered in terms of lightness, chroma and hue. Each of the images was displayed on a 2-inch QVGA mobile LCD and assessed by a panel of 10 observers in terms of image quality using a categorical judgment method. Only colour attribute modeling was carried out in this paper. Image statistical measures were established to quantify the image quality of natural colour images varying in colour. Those were memory colour reproduction ratio (MCRR), mean chroma and 95 percentile lightness. The Pearson correlation between the model predictions and their corresponding psychophysical data was 0.88 and the coefficient of variance was 18. The model outperformed observer accuracy in terms of those two measures. It is also significant that the subjective scale of image quality was bridged with objective metrics such as the image statistical measures. Introduction Image quality has been recognised as one of the top considerations in the display manufacturing industry, where there is a perpetual trade-off between quality and cost. Hence, an image quality metric, which can accurately quantify the quality of an image according to human visual perception, is strongly desired. Objective evaluation involves physical measurement of images but generally fails to consider human visual characteristics. Therefore, psychophysical experimental results are required for developing metrics. Subjective image quality research can be divided into two major approaches: external (or relative image quality) and internal reference (or absolute image quality). The former assumes that the image quality of reproductions corresponds to perceptible visual difference from its original. A number of these metrics have been suggested and widely used such as CIELAB colour difference equation and CIEDE2000. 6 SCIELAB 18 was developed in 1996 as an image difference metric accounting for image spatial properties. Internal reference image quality is defined as the image quality of a given image corresponding to perceptible visual difference from its memory prototype. The category judgment method is appropriate for this approach, in which observers assess a single image by perceptual comparison with a cognitively represented reference, for which the original is not presented. 9 There has been some effort to appraise an image without an original based on information theory and the similarity to the memory colours of sky, grass, and Caucasian skin. It was found that the appearance of particular memory colours are remembered slightly differently from the colour measurement from the real world. Previous Study In the previous study, affective attributes in image quality modelling were investigated. These included naturalness and clearness. The experiment was originally designed to develop two types of image quality models: perceptual (or high-level) and physical (or low-level). The former was studied in the previous study and represents an image quality model that involves some perceptual attributes as input values such as naturalness, clearness or sharpness. Five natural test images were rendered in terms of seven physical parameters: two types of lightness rendering functions, chroma, hue, peak-white luminance, resolution, bit depth and correlated colour temperature (CCT). Observers used 9 categories (1 to 9) to appraise each image, according to 7 perceptual attributes: naturalness, clearness, sharpness, contrast, colourfulness, image quality and preference. The attributes were inter-compared using Pearson correlation. Naturalness and image quality showed a very high correlation (0.96). Clearness seemed to be highly associated with sharpness (0.97). Image quality and preference were judged to be the same attribute (0.99). Image quality was modelled by the 5 perceptual attributes through a stepwise regression method, and it was found that naturalness and clearness are the principal affective attributes in the image quality of mobile displays. The second image quality model is based on the physical attributes which include image statistical measures in the colour, spatial, or temporal domains. In the current paper, only colour attributes were considered and the accumulated mean opinion score (MOS) values of image quality from the previous study were used to develop an image colour-quality model based on image statistical measures such as memory colour reproduction, mean chroma and 95 percentile lightness. The spatial attributes were left for future research.