Computational Colour Approach to Digital Imaging, Human Perception, Computer Vision and AR/VR/MR

点击翻译按钮获取中文摘要

Computational Colour Approach to Digital Imaging, Human Perception, Computer Vision and AR/VR/MR This proposal is about how the colours of objects are perceived by different `observers'-different people, different cameras-and under different lights (daylight, LED, fluorescent, tungsten). Colour is a very interdisciplinary field touching on psychology, philosophy, chemistry, physics, and computer science. My approach to understanding colour is to view colour perception as a computational process. As such, the models of colour perception developed in my laboratory are formulated as algorithms that can be tested both in terms of whether or not they simply provide the expected results, and secondly in terms of whether they operate in a way that is congruent with what is known about human colour perception from the psychophysical experiments conducted by psychologists. The fundamental difficulty in understanding and modeling colour perception is that because humans have only 3 types of colour-sensitive cones there is no one-to-one correspondence between the wavelengths of light entering the eye and perceived colour. There are many other difficulties too, such as how the same reflected-light spectrum may look different in different contexts, and these all contribute to colour being a fascinating field of research. These difficulties, however, present problems for the digital camera industry, the digital printing industry, the digital display industry, the textile industry, the lighting industry, the scientific use of colour (e.g., in medical applications), the digital preservation of artwork, and most recently to colour accuracy in augmented/mixed reality applications. The research projects in this proposal address fundamental issues of colour science that have direct application to all these technology areas. The objectives of the proposed research build on the recent progress in my laboratory that include work on metamerism, colour constancy and the limits thereof, establishing the set of all theoretically-possible colours, evaluating the colour rendering properties of lights, and evaluating the colour accuracy of digital cameras. The proposed objectives include: (i) a new illuminant-invariant approach to object classification and recognition with implications for machine learning generally; (ii) building a dataset with accurate ground truth of multispectral images of scenes containing complex, spatially-varying lighting for use by the colour research community; (iii) providing colorimetrically and perceptually accurate colour for augmented reality applications; (iv) further developing a new theory of how colour discrimination varies with hue; and (v) understanding human vision in terms of compressive sensing. The research budget is chiefly for the training of highly qualified personnel; namely, support of student salaries and their conference travel. My previous students now all have successful careers in the colour-imaging field.