Algorithms for colour image processing based on neurological models

373 p. : il., gráf., fot., tablas

Colour image processing is nowadays mostly achieved through the extrapolation of algorithms developed for images in grey levels into three colour planes, either RGB or some transformed planes, such as HSI, CIELAB... These techniques provide reliable solutions only in simple situations. As colour is a perception and not a characteristic inherent to objects, this thesis has developed new bioinspired algorithms for colour image processing. The work of this thesis has joined elements in colour theory and processing undertaken in the human visual system. A new functional model of the retina has been developed where each cell type has been characterised according to its connections, distribution and size. A retina architecture has been created which provides detailed information about its cell elements and organisation. This has allowed the creation of a retina model that generates a set of parallel output channels as happens in the human retina. The level of detail provided in the model has allowed the characterisation of each of the pathways with a precision that is not present in existing models described in scientific publications. The development of a colour processing model requires the combination of a functional retina model with colour appearance models. This union has achieved a new algorithm for colour image processing that provides colour attributes, such as: hue, lightness, brightness, saturation, chroma, colourfulness as well as edge detection components both in chromatic as well as achromatic components. The results provided by this model have been compared with CIECAM02 model's ones and have obtained noticeably better results in the "ab" plane and in the attributes calculated on Munsell colour samples. The colour processing model is backed by its results and has allowed identifying output channels of the retina that make up the usual "a", "b" and "A" channels in colour appearance models. This model entails a step forward on colour processing techniques that shall be of great use for image segmentation, characterisation and object identification. Key Words Colour image processing, neuroinspired models, computational modelling, colour appearance models. Colour image processing is nowadays mostly achieved through the extrapolation of algorithms developed for images in grey levels into three colour planes, either RGB or some transformed planes, such as HSI, CIELAB... These techniques provide reliable solutions only in simple situations. As colour is a perception and not a characteristic inherent to objects, this thesis has developed new bioinspired algorithms for colour image processing. The work of this thesis has joined elements in colour theory and processing undertaken in the human visual system. A new functional model of the retina has been developed where each cell type has been characterised according to its connections, distribution and size. A retina architecture has been created which provides detailed information about its cell elements and organisation. This has allowed the creation of a retina model that generates a set of parallel output channels as happens in the human retina. The level of detail provided in the model has allowed the characterisation of each of the pathways with a precision that is not present in existing models described in scientific publications. The development of a colour processing model requires the combination of a functional retina model with colour appearance models. This union has achieved a new algorithm for colour image processing that provides colour attributes, such as: hue, lightness, brightness, saturation, chroma, colourfulness as well as edge detection components both in chromatic as well as achromatic components. The results provided by this model have been compared with CIECAM02 model's ones and have obtained noticeably better results in the "ab" plane and in the attributes calculated on Munsell colour samples. The colour processing model is backed by its results and has allowed identifying output channels of the retina that make up the usual "a", "b" and "A" channels in colour appearance models. This model entails a step forward on colour processing techniques that shall be of great use for image segmentation, characterisation and object identification. Key Words - Colour image processing, neuroinspired models, computational modelling, colour appearance models.