Human Vision and Electronic Imaging XIV

Single-Scale SSIM (SS-SSIM) and Multi-Scale SSIM (MS-SSIM), objective metrics introduced by Wang and Bovik, have shown a good correlation with perceived quality. Therefore, SS-SSIM and MS-SSIM have a big potential to also become the perceptual metrics for measuring perceived quality in Digital Cinema applications.

Our goal is to design SS-SSIM and MS-SSIM metrics with input parameters that take into account the Digital Cinema source material characteristics (resolution, dynamic range, frame rate / motion, content) and viewing conditions. These metrics are then utilized to measure the perceived quality of high quality digital imagery. To validate and to confirm the results, these are compared with the PSNR and with a subjective evaluation/assessment carried out by human observers in a DCI specified movie theater environment.

In a series of experiments, we collected subjective judgments of image harmony and other overall perceived attributes for images representative of typical consumer photography. The ratings were used to reveal underlying statistical properties and semantic factors that can explain peoples\u2019 perception of harmony. We also analyzed a much larger database of images to determine patterns of color combinations and spatial configuration with the probabilities of their occurrences to hypothesize whether the characteristics representing typical living environments affect individual harmony response.
The results demonstrate the potential of this approach for understanding visual harmony and aesthetic appeal.

In this paper we make two contributions. First, on the algorithm side we reconsider the Socolinsky and Wolff algorithm for colour to greyscale conversion . This algorithm, which is the most mathematically elegant, often scored well in preference experiments but occasionally it introduces artefacts which spoil the appearance of the final image. These artefacts are intrinsic to the method and stem from the underlying approach which computes a grey scale image by a) calculating approximate luminance-type derivatives for the colour image and b) reintegrating these to obtain a greyscale image. Unfortunately, the sign of the derivative vector is occasionally unknown and, in the current theory, is set arbitrarily. However, choosing the wrong sign can lead to unnatural contrast gradients (not apparent in the colour original). Our first contribution is to show how this sign problem can be ameliorated using a generalised definition of luminance and the Markov Random Field theory.

In the second part of this paper we will report on a large set of preference experiments which involve more images and more algorithms than reported in CIC 15. We will also pay particular attention to testing our modified Socolinsky and Wolff algorithm. Based on initial results we expect improved preference scores.

Moreover, within a real-world scene, light incident on a measurement surface is composed of a direct and an indirect part. Neumann and Schanda have shown for the cube model that interreflections can influence the CRI of an illuminant. We analyze how color rendering indices vary in a real-world scene with mixed direct and indirect illumination and recommend the usage of a spectral rendering engine instead of an RGB based renderer for reasons of accuracy of CRI calculations.

We are interested in the co-location statistics of edge discontinuities in corresponding luminance and disparity images, a topic that has not been studied much but which may prove quite useful. We designed a Laplacian of Gaussian (LoG) filterbank with center frequencies roughly matching human contrast sensitivity. The LoG filters were used to detect edges in the grayscale intensity images, the co-registered range maps, and the derived disparity maps.

At all detected luminance, disparity, and range edges, we studied the pair-wise linear correlation among luminance gradient, disparity gradient, and range gradient. We found that the strongest correlation (about 0.37) exists between luminance gradient and range gradient, yet the linear correlation between luminance gradient and disparity gradient was low. The correlation between luminance gradient and range gradient indicates the possibility of luminance gradient being used by the visual system in absolute distance estimation.