Supervisor: Dr. Philipp Urban
Looking at pictures while browsing the internet, you might have already found yourself thinking “The quality of this picture is pretty bad!”. But what does this mean really? And, how does one automatically predict this kind of judgment? Assessing the quality of an image with respect to human perception pertains to predicting how our visual system interprets features such as contrast, structure, chroma or hue. There is a tremendous amount of literature on the topic for gray-scale and colour image quality assessment. Most of these studies rely on the assumption that the image whose quality needs to be assessed is shown to a standard observer as a single stimulus, on a low-dynamic range, calibrated colour display. Additionally, in the case of full-reference image quality assessment, an original, non-distorted image is also available for immediate comparison.
Recently, a need to account for more complex and/or diverse viewing conditions grew with the advent of multispectral technologies (which allow to render a scene for different viewing conditions), multi-channel printing (which aim is to control the reflectance of the print, instead of just its colour under daylight) and high-dynamic range technologies (which allow to simultaneously capture details in dark and bright tones).
The aim of this work package within the CP7.0 project was primarily to understand the key challenges in Spectral Image Quality Assessment (S-IQA) for printing applications. For instance, a typical image processing technique used in printing workflows is gamut mapping. It consists of making use of a limited set of printable colours to represent the image with as little quality loss as possible. If the image is described in a wide gamut such as that of the Adobe RGB colorspace, it may then need to undergo drastic changes as only a fraction of this gamut can indeed be printed. Needless to say, quality control is of the essence in such processing. The notion of quality, though subject to different interpretations, should be in this case linked to that of perception, i.e. human judgment should be considered the reference measure of an image’s quality. While human perception of colour images has been extensively studied, that of multispectral images required further investigations. We dedicated our efforts to finding perceptual redundancies in renderings of the same multispectral scene for different illuminants, and found for instance that our perception of lightness attributes (including lightness-structure and -contrast) is significantly less dependent on the scene illumination than that of chromatic attributes such as hue and chroma, even when considering the ability of our visual system to compensate for different illuminants (chromatic adaptation). This allowed us to develop the first perception-driven S-IQA measure, referred to as the Spectral Image Difference (SID). As well, we developed a low-dimensional representation of multispectral data coined LabAB to enable their efficient (i.e. computationally inexpensive) treatment.
We also studied High-Dynamic Range Image Quality Assessment (HDR-IQA), and found that a single working colorspace can be used to represent both Low- and High-Dynamic Range data so as to compare them. This enables for instance the direct comparison, in terms of perception, of an original and tone-mapped images. The challenge here lies in correctly extracting the perceived information from the HDR image, which pertains essentially to finding an accurate model for luminance adaptation. Indeed, we can only perceive a limited dynamic range of luminances at a time, which compels us to adapt in order to be able to deal with the variety of luminance ranges in the nature (e.g. day vision versus night vision). The range of HDR images is often larger than that we can perceive simultaneously, which therefore induces luminance adaptation and implies that some details of the image will not be perceived. Consequently, one needs to understand what exactly is seen from an HDR image (on an HDR display) for quality assessment purposes.
In conclusion, this project led us to study the ability of our visual system to adapt to different viewing conditions and to account for this ability in the development of new image quality models.
Publication and Dissemination
- Blahova, J., LeMoan, S. and Urban, P. (2013), The impact of illumination on the perceived quality of spectral reproductions, Color Image Processing Workshop, Berlin, Germany.
- Le Moan, S. and Urban, P. (2013a), Evaluating the perceived quality of spectral images, IEEE International Conference on Image Processing, Melbourne, Australia.
- Le Moan, S. and Urban, P. (2013b), Image quality and change of illuminant: An infor- mation theoretic evaluation, in ‘Color and Imaging Conference’, number CIC21, Society for Imaging Science and Technology, pp. 102 − 107.
- Le Moan, S. and Urban, P., (2014) Image-Difference Prediction: From Color to Spectral, Transactions on Image Processing, vol. 23, no. 5, pp. 2058-2068,IEEE.
- Le Moan, S. and Urban, P. (2014), A new connection space for low-dimensional spectral color management, SPIE Electronic Imaging, San Francisco, CA, USA.
- Coppel, L.G., Le Moan, S., Zˇitinski, P. E., Slavuj, R., Hardeberg, J. Y. (2014), Next generation printing – Towards spectral proofing, ‘Advances in Printing and Media Tech- nology, Print and Media Research for the Benefit of Industry and Society’, IARIGAI print and media research, Swansea, UK.
- Le Moan, S. and Urban, P., (2014), Spectral printing with a CMYKRGB printer: a closer look, 22nd Color and Imaging Conference, Boston, MS, USA, November 2014, IS&T.
- Le Moan, S. and Urban, P. (2015), Evaluating the Multi-Scale iCID Metric, SPIE Elec- tronic Imaging, San Francisco, CA, USA.
- Le Moan, S., George S., Pedersen, M., Blahova, J., Hardeberg, J.Y. (2015), A database for spectral image quality, (submitted) to SPIE Electronic Imaging, San Francisco, CA, USA.
- Le Moan, S. and Coppel, L. G. (2015) Perceived Quality of Printed Images on Fluo- rescing Substrates under Various Illuminations, in proceedings of the 16th International Symposium on Multispectral Color Science, AIC 2015 Mid-term meeting, Tokyo, May 2015, Color Science Association of Japan.