Computational colour vision

点击翻译按钮获取中文摘要

The overall objective of this research is to fully understand the computational processes underlying colour perception. While of intrinsic scientific interest, understanding colour perception is also essential for improving colour reproduction in the context of digital cameras, displays, and printers. In addition, it has important applications in computer vision in terms of colour-based object recognition, and the analysis of the colour changes created by shadows, interreflections, wetness and transparency. Our approach to these issues utilizes the computational view of colour; i.e., that human colour perception and the ways in which colour is used in understanding the visual world can be explored and explained as computational processes. Specific short-term objectives include: (1) Working on full-spectrum printing to overcome the problem that arises with traditional 4-colour CMYK printing; namely, that printed colours that match one another under one light can, under a second light, change their appearance such that they no longer match. (2) Investigating whether colour images with more than the standard 3 RGB bands might lead to improved colour constancy. The interesting and counterintuitive result we have so far is that the extra information in multispectral images does not yield significant improvement. (3) Improving picture rendering by identifying the manipulations that are predicted by colour appearance theory and compare those to the manipulations that have been practiced for decades in photography, video, and motion picture production. The aim is to formalize some of the 'art' of film making techniques in terms of existing colour appearance models and simultaneously explore the limitations of these models. (4) Exploring uses of quaternions (4-tuple complex numbers) as a representation for RGB colours. The quaternion representation has the advantage that the 3 RGB components are treated as a single number.