Our approach aims to go one step further in the understanding of the image content and, under the assumption that different subject categories require different composition techniques, introduces an additional scene type classification step, which, combined with the use of state of the art feature sets, should yield a significant improvement over the current performance results.

The main goal of this project is to find a way of quantifying smoothness of color transforms through analysis of LUTs based device characterization process and the factors which affect on it and test different image quality metrics. It requires conducting a number of experiments for determining and estimating thresholds of factors for predicting unsmooth transform and corresponding human perceptual evaluation of smoothness of color transforms in comparison with quantitative evaluation results provided by different image quality metrics. The evaluation smoothness of LUTs-based color transforms will allow to avoid undesirable results in image color reproduction systems such as artifacts and distortions of image content and improving device characterization process for achieving smooth color transform. It is an attempt to continue work in direction of getting qualitative reproduction of color images fast and economically and developing technology advances for obtaining it.

To get these reflectances of Neugebauer Primaries one has to find out the way to print out them so that he/she could measure the reflectances. It has the following problems. The number of patches we are going to print increases exponentially with the number of colorants of the printer since the number of possible combination of colorants is given by 2n , where n is the number of colorants of the printer. For Spectral printer with more than 7 or 8 colorants, printing all those Neugebauer Primaries and measure them will consume lots of time and materials. The other problem is that finding out the way to really print out all possible combinations is very difficult. There are some charts which incorporate all Neugebauer Primaries of CMYK printer. The problem will begin when we consider printers with more colorants than these 4. There are no charts which actually incorporate Neugebauer Primaries of 8 channel printer. In this case the researcher forced to have the SDK of the printer and find out some way to order the printer to print anything he/she wants.

In this thesis we tested Kubelka-Monk theory in order to estimate those Neugebauer Primaries so that we could save our resources, power and time. We used different kinds of papers for test and for recommending more cheapest and accurate way of using this theory. Then we spectrally model the CMYK HP Deskjet 1220C printer and Xerox phaser 7760 laser printer using spectral Neugebauer model and Yule-Nielsen modified spectral Neugebauer (YNSN) model. Using these models we show that how much more improvements we could actually get by estimating Neugebauer Primaries using Kubelka-monk theory rather than using real measurements of the Neugebauer Primaries.

Multispectral imaging systems remedy the problems of conventional three channel (RGB) color imaging like metamerism and dependency on acquired conditions and at the same time presents high spatial and spectral resolution. A multispectral image is composed of several monochannel images of the same object, each image holds data about a specific wavelength depending on the filter used. The major problems of existing multispectral imaging systems are that they are slow as they require multiple takes and/or they are quite expensive, contributing to the current lack of widespread use in the consumer segment. This thesis will explore creating a fast, practicable and affordable multispectral system with the use of two commercially available digital cameras. Each camera is equipped with an optical filter. These two filters are chosen so that they modify and spread the sensitivities of the cameras so that they become well spaced throughout the visible spectrum covering complementary wave-bands, thus giving rise to six channel multispectral system.

Like multispectral imaging, 3D imaging systems are also gaining popularity and are of great use in imaging fields. It would be great advantage if we can have an integrated system capable of both multispectral and at the same time 3D imaging. This thesis, therefore, aims at this in conceiving such a multispectral-stereo system. The two cameras modified with appropriate filters that forms six channel multispectral system are used in stereoscopic configurations to acquire depth information making it capable of 3D imaging as well. This leads to a faster and practicable and at the same time affordable multispectral-stereo systems.

Such a system could be used for many applications, for example for 3D artwork object acquisition. Knowing the spectral reflectance allows us to simulate the appearance of a 3D object under any virtual illuminant. Moreover, it lets us store this valuable information for future restoration.

In the first part of this thesis we consider pointwise display color characterization. We propose,evaluate and improve several methods to control the color in displays.

We investigate deeply the PLVC (Piecewise Linear assuming Variation in Chromaticity) model especially in comparison to the PLCC (Piecewise Linear assuming Chromaticity Constancy) model. We show that this model can be highly beneficial for LCD (LiquidCrystal Display) technology. We evaluate and improve a end-user method proposed by Bala and Braun. This method is quick and simple and does not need any measurement device other than a simple digital color camera. We confirm that this method gives significantly better results than using default gamma settings for both LCD and DLP (Digital Light Processing) projectors.

We focus on the distribution of color patches in color space for the establishment of 3D LUT (Look Up Table)models. We propose a new accurate display color characterization model based on polyharmonic spline interpolation. Thismodel shows good results and is applied in real time for the accurate colorimetric rendering of multi-spectral images of art paintings viewed under virtual illuminants. We propose methods to build an optimized structure that permits to invert any display color characterization forward model. Several criteria linked with the grid itself or with an evaluation data set are tested. Our evaluation shows that in using our methods, we can achieve better results than with a regular equidistributed grid.

In a second part, we establish a basis for spatial color characterization via the quantitativeanalysis of the color shift and its spatial variation throughout the display area. We show thatthe spatial chromaticity shift is not negligible in some cases and that some features are spatiallyinvariant within one display of a given technology

Two research questions have been formulated as the basis for this thesis:

• How can we improve existing image difference metrics ?

• What is the performance of the Difference of Gaussians model in image difference metrics ?

A first test on a set of gamut mapped images will give an idea of the performance in correlation of the two metrics. A second experiment will be performed on images with single and multiple variations of contrast, lightness, and saturation. The performance in correlation will be given using data from a psychophysical experiment. Results will show that the proposed metrics have a low performance on the dataset of gamut mapped images and they do not seem to be appropriate for the second dataset as well.

We will also demonstrate through the experiments the importance of spatial filtering for color image difference metrics, and also that the configuration of these two metrics should change according to the type of distorted images to ensure better performance.

This research project investigates the importance of region-of-interest on image difference metrics. Region-of-interest has been extracted by using an eye tracker, but also by manual marking by the observers. 3 different tasks were performed by the observers while their gaze position was recorded. Further a manual marking of region-of-interest together with a questionnaire to map background knowledge was carried out. The information found on how we perceive and examine images has been applied to different image difference metrics, such as deltaEab, S-CIELAB, iCAM, SSIM and the hue angle algorithm. The issues regarding how observers look at images given different tasks are also discussed and analyzed.

The results indicate that region-of-interest improves image difference metrics, especially when the metrics already have a low performance in term of linear correlation between perceived and calculated difference. There are no clear evident that one type of region-of-interest outperform other types. The improvement in performance is therefore both scene and metric dependent.

Results also show that observers have different areas of attention according the task given to them, as freeview, rating image difference and marking important regions. The common denominator within every task is faces, and this is clearly important in all tasks for the observers. Within areas of attention will change whether the observer is an expert or non-expert.

The iris recognition process usually consists of four major steps. The first step is to segment the iris out of the image containing the eye and part of the face, which localizes the iris pattern. Step two is the normalization, here the iris pattern will be extracted and scaled to a predefined size. Step tree is the encoding phase, here the details of the iris are filtered, extracted and represented in an iris code. The last step is the comparison, where two iris codes will be compared and a similarity score is computed. In this thesis we have focused on the encoding and filter algorithms, in step tree, under different unfavorable conditions. We used the open source code of Libor Masek and extend it with different filtering algorithms. The filters which were included in the analysis are: two Haar filter, one Log-Gabor filter and one Laplacian of Gaussian filter; which generate iris codes with sizes 702 bit, 87 bit, 9600 bit and 9600 bit respectively. The filtering algorithms have been tested using a database of 500 iris images. The images in the database have been corrupted using different degradation models. We used four different degradation models: additive Gaussian noise and blur, changing the light intensity and rotating the images. Two performance measures were used: the False Acceptance Rate (FAR) and False Rejection Rate (FRR). The first is estimated using inter-class comparisons while the second is estimated using intra-class comparisons. The total number of comparisons performed in the experiments are approximately seven million comparisons.

Based on the experimental results we obtained in this work, we can conclude that the performance of all the tested algorithms is dramatically affected by the degradations. The major cause of the performance drop is the sensitivity of the segmentation process to such degradations. These degradations introduce segmentation errors in many of the images.

The experimental results also show that the Log-Gabor and Laplacian of Gaussian filters are best under optimal conditions. Under non-optimal conditions however, the Haar filter with 702 bit representation achieves close to best result under most of the degradation models. This maybe due to the fact that it extracts the iris characteristics based on fewer but most prominent details in the iris which are relatively more robust to degradations.

for comparing images on a computer monitor. The goal of this comparison is to determine if

pictures, which correspond to specific image data processing algorithms, have significant

difference in quality. If the difference is noticed, rank of algorithms can be formed and the best

algorithm selected. The numerous tests can reveal that one specific method is frequently first in a

row. If such situation is encountered, this algorithm can be set as an industrial standard for image

data processing which matches specific requirements. Then users, seeing the obvious leader in

particular industry section, can more easily decide and avoid ambiguity when choosing the most

suitable solution meeting their needs.

The software functionality has to include the most important features expected. The

application should support different kinds of tests (paired comparison, category judgement) for a

variety of tests to perform and to achieve more accurate final results. It would be convenient to

have a module for administration (users, images, etc.), which simplifies the researches work during

the test period. The software also must have a module for simple data analysis according to

standardized statistical methods. It must be possible to export the empirical data in a suitable

format for compatibility with other wide-used software. The application for performing image

quality tests was called QuickEval, which stands for “quick evaluation”.

Viewing statistical results just like a string of numbers don’t give the expected

picturesqueness for the researcher. Seeking for clearer evaluation and to present the results in a

more suitable format for human’s eye, we were suggested to create a module for drawing charts

based on data acquired by the comparison module. This improved overall functionality of the

developed software application and gave even more ideas for further expandability. The module

for drawing graphics has a name of ChartDrawer and this explains its function, obviously.