The title of the thesis focuses the attention on hyperspectral image segmentation, that is, we want to detect salient regions in a hyperspectral image and isolate them as accurate as possible. This purpose presents two main problems: Firstly, the fact of using hyperspectral imaging not only give us a huge amount of information, but we also have to face the problem of selecting somehow the information avoiding redundancies.Secondly, the problem of segmentation strictly speaking is still a challenging question whatever the input image would be.This thesis is focused on solving the whole process by means of building an image processing method that analyses and optimises the information acquired by a multispectral device. After that, it detects the main regions that are present in the scene in an image segmentation procedure. Therefore, this work will be divided into two parts. In the first part, an approach for selecting the most relevant subset of input bands will be presented. In the second part, this reduced representation of the initial bands will be the input data of a segmentation method.Finally, the main contributions of this PhD work could be briefly summarised as follows. On the one hand, we have proposed a pre-processing stage with an unsupervised band selection approach based on information measures that reduces considerably the amount of data. This approach has been successfully compared with well-known algorithms of the literature, showing its good performance with regard to pixel image classification tasks. On the other hand, after the band selection stage, two unsupervised segmentation procedures for detecting the main parts in multispectral images have been also developed. Regarding to this segmentation part, we have mainly contributed with two measures of similarity among regions. An objective functional for selecting an optimal (or close to optimal) partition of the image is another relevant contribution too.

The annotated papers that are submitted as part of this thesis consider the phenomenon of colour at the fundamental, technical, and application levels, and they were written and published by Oulton between 1990 and 2009. The papers disclose significant insights by the author into colorimetric modeling theory and report aspects of the author's work that have led to commercially successful practical applications. The academic significance of these papers is evident in their citation record; their practical value is shown by a number of successful industrial collaboration programmes, and through the award of national prizes for innovation by the Worshipful Company of Dyers, and the Society of Dyers and Colorists. The published research primarily concerns digital devices that either capture or reproduce coloured images. For example, the research problem of how to calibrate the colour on computer CRT screens, which was thought at the time to be intractable, was reported by Oulton in paper 1 to be solved at the two to three significant figure level of colorimetric accuracy. This world leading level of accuracy was subsequently confirmed using a comprehensive data set in paper 7, and has been exploited internationally in commercial computer aided design and colour communication systems by Textile Computer Systems Ltd and Datacolor Inc. Further research problems resolved by Oulton in the presented papers include how to predict the colorimetric sensitivity of dye recipes; how to design, test, and fine-tune the spectral response of digital cameras; and how the individual customers in a shop can be tracked automatically to reveal their buying behavior, using coloured CCTV images.The challenge to the standard CIE colorimetric model posed by the results of Dr W.A. Thornton was analyzed and satisfactorily explained by Oulton in papers 2, 3 and 4. It is also shown that Thornton's results do not in any way compromise either the practice of colorimetry based on the CIE Standard Observer, or the validity of its quantifying data sets. It is also additionally shown under the annotation of paper 4 presented here, that the success of the CIE colorimetric model has a clearly demonstrable theoretical basis.In all but one of the presented papers the convention is maintained that the standard CIE XYZ co-ordinate model should be used as the reference basis, when modeling the properties of colour and quantifying its uses. The final paper to be published (and presented here as paper 4) challenges this convention and demonstrates that a context free and formally defined alternative reference basis may be used in colorimetric modeling with significant advantage. It is also shown in paper 4 that under the specified axioms, any cross dependency that is potentially non linear can in principle be resolved into its component scalar and additive relationships, and that the causes of scalar non linearity may be characterized independently from the causes of linearly additive cross dependency. The result is a widely applicable analytical and experimental design method for resolving complex cross dependent relationships in general and in particular, for resolving those between the spectral visual stimuli and the psychophysical response to them.

This project deals with automatic recognition of facial expression in colour pictures. At first, the colour-based face detection is accomplished, three colour spaces are used: RGB, HSV and YCbcCr. As next, the pictures are automatically cropped so that only the face region is present. It is accomplished by computing the borders of the face region based on knowledge of position of eyes, nose and mouth. From the face region, the feature vector is obtained using a bank of Gabor filters. The project introduces two different kinds of Gabor filters and proposes a new bank of filters. The feature vector is used as an input to the neural network. The neural network was trained on a set of pictures from AR database created for facial expression recognition. The output of the network is the facial expression the input picture was assigned to. This project mentions the testing for different settings of the neural network and presents and discuss the recognition results of the network.

This master’s thesis is focused on automating the process of differential leukocyte count in peripherial blood using image processing. It deals with the design of the processing of digital images - from scanning and image preprocessing, segmentation nucleus and cytoplasm, feature selection and classifier, including testing on a set of images that were scanned in the context of this work. This work introduces used segmentation methods and classification procedures which separate nucleus and the cytoplasm of leukocytes. A statistical analysis is performed on the basis of these structures. Following adequate statistical parameters, a set of features has been chosen. This data then go through a classification process realized by three artificial neural networks. Overall were classified 5 types of leukocytes: neutropfiles, lymphocytes, monocytes, eosinophiles and basophiles. The sensitivity and specificity of the classification made for 4 out of 5 leukocyte types (neutropfiles, lymphocytes, monocytes, eosinophiles) is higher than 90 %. Sensitivity of classiffication basophiles was evaluated at 75 % and specificity at 67 %. The total ability of classification has been tested on 111 leukocytes and was approximately 91% successful. All algorithms were created in the MATLAB program.