ProLAB: Perceptually Uniform Projective Colour Coordinates System

Aiming at advancing the fundamental science, Smart Engines R&D team is closely working with the research group from the Russian Academy of Sciences — specifically the Institute for Information Transmission Problems. Our collaborative research has recently resulted in the development of the new colour coordinate system. The system, in our view, is the most suitable framework for linear colour analysis, and is given further elaboration below.

Recently, Smart Engines researchers together with the team from the Institute for Information Transmission Problems RAS, in the scientific collaboration devoted to the advancement of colour vision science have developed the new colour coordinate system. That system is claimed to be the best suitable framework for linear color analysis. The results of the research have been presented by Smart Engines at the 25th Symposium of the International Color Vision Society (ICVS). Established in 1969 in Sweden, ICVS is a large R&D group of physiologists, psychologists, physicists, geneticists, optometrists, ophthalmologists and visual scientists, which studies various aspects of colour vision and colour vision deficiencies. Held in Riga, Latvia, earlier this summer, the event brought together 167 participants from 19 countries.

Below, you will find some details of the invention, including the poster which was presented at the conference.

Aiming at advancing the fundamental science, Smart Engines R&D team is closely working with the research group from the Russian Academy of Sciences — specifically the Institute for Information Transmission Problems. Our collaborative research has recently resulted in the development of the new colour coordinate system. The system, in our view, is the most suitable framework for linear colour analysis, and is given further elaboration below.

Various algorithms of technical colour vision (mostly related to colour constancy problem) operate in so called linear colour spaces which are obtained by linear mappings of the spectral power distribution space. This practice is motivated by the linearity of the underlying optical image formation model. However, requirement of mapping linearity is redundant since affine properties of manifolds are preserved not only by linear transformations. On the other hand, linear regression in linear colour spaces (hardware RGB or CIE XYZ) is unreasonable, because euclidean distance in such spaces does not approximate the colour difference well. This holds both for perceptual difference and deviations caused by heteroscedastic optical noise.

The researchers from Smart Engines and IITP RAS have proposed the ProLAB colour space obtained as a projective mapping of the spectral power distribution space which is more adequate for colour image analysis algorithms. The Euclidean distance in the ProLAB coordinate space provides a closer approximation of perceptual differences between colours in comparison with the widely used CIE LAB. Also, the transition into ProLAB coordinates makes image noise more homoscedastic. For now it looks to be the best suitable framework for linear colour analysis in computer vision.

A software development company with a broad expertise in computer vision, document processing and OCR, mobile recognition systems and AI. www.smartengines.com