The quality of a digital image pipeline relies greatly on its color reproduction which should at a minimum handle the color constancy, and the final judgment of the excellence of the pipeline is made through subjective observations by humans. This dissertation addresses a few topics surrounding the color processing of digital image pipelines from a practical point of view. Color processing fundamentals will be discussed in the beginning to form a background understanding for the topics that follow.A memory color assisted illuminant estimation algorithm is then introduced after a review of memory colors and some modeling techniques. Spectral sensitivity of the camera is required by many color constancy algorithms but such data is often not readily available. To tackle this problem, an alternative method to the spectral characterization for color constancy parameter calibration is proposed. Hue control in color reproduction can be of great importance especially when memory colors are concerned. A hue constrained matrix optimization algorithm is introduced to address this issue, followed by a psychophysical study to systematically arrive at a recommendation for the optimized preferred color reproduction. At the end, a color constancy algorithm for high dynamic range scenes observing multiple illuminants is proposed. / Graduation date: 2012

Le tatouage invisible d’images d’identité imprimées sur un support en plastique est un problème difficile qui intéresse le monde industriel. Dans cette étude, nous avons développé un algorithme de tatouage robuste aux diverses attaques présentes dans ce cas. Ces attaques sont liées aux processus d’impression/numérisation sur le support plastique ainsi qu’aux dégradations qu’une carte plastique peut rencontrer le long de sa durée de vie. La méthode de tatouage opère dans le domaine de Fourier car cette transformée présente des propriétés d’invariances aux attaques géométriques globales. Une méthode préventive consiste en un prétraitement de l’image originale avant le processus d’insertion qui réduit la variance du vecteur support de la marque. Une méthode corrective comporte deux contre-attaques corrigeant le flou et les variations colorimétriques. Pour une probabilité de fausse alarme de 10⁻⁴, nous avons obtenu une amélioration moyenne de 22% par rapport à la méthode de référence lorsque seule la méthode préventive est utilisée. La combinaison de la méthode préventive avec la méthode corrective correspond à un taux de détection supérieur à 99%. L’algorithme de détection prends moins de 1 seconde pour à une image de 512×512 pixels avec un ordinateur classique ce qui est compatible avec l’application industrielle visée. / Invisible watermarking for ID images printed on plastic card support is a challenging problem that interests the industrial world. In this study, we developed a watermarking algorithm robust to various attacks present in this case. These attacks are mainly related to the print/scan process on the plastic support and the degradations that an ID card can encounter along its lifetime. The watermarking scheme operates in the Fourier domain as this transform has invariance properties against global geometrical transformations. A preventive method consists of pre-processing the host image before the embedding process that reduces the variance of the embeddable vector. A curative method comprises two counterattacks dealing with blurring and color variations. For a false alarm probability of 10⁻⁴, we obtained an average improvement of 22% over the reference method when only preventative method is used. The combination of the preventive and curative methods leads to a detection rate greater than 99%. The detection algorithm takes less than 1 second for a 512×512 image with a conventional computer, which is compatible with the industrial application in question.

Color image acquisition and image processing have become a key in modern data application. In order to provide high quality images, the field of accurate acquisition is most important in respect to all further processing steps. But a whole variety of current image sensors possess incorrect color rendition due to insufficient accuracy of optical sensor parameters. This is detrimental especially for color sensors, because in these cases specific color information will be incorrectly acquired. Further, traditional color correction methods do not use information on the specific sensor spectral sensitivity, thus losing substantial information for color correction. The problem is investigated by introducing an algorithmic correction method which is capable of correcting dysfunctional sensor properties. The correction method is based on an enhancement of the CIE color perception model. According to this, color perception is modelled as a special integral transformation, where the spectral sensitivities of the photo receptors represent the base functions of the transformation. It is shown that different sets of photo receptors show the same perception, when their spectral sensitivities are linear dependent. On the other hand, photo receptors with no linear dependency show different perception and there is no analytical transformation between them. Thus, a perfect color correction is only possible if photo sensor and human perception show a linear dependency. In case of dissentient sensor characteristics, the correction method of spectral reconstruction can determine an optimal solution using a least square error optimization. Applying sensors with more than three color channels, this correction method can show improved results due to a better approximation. For implementation of the color correction scheme, different sensor designs have been developed. Compared with currently dominating CCD (Charge Coupled Device) technology, a realisation of image sensors based on CMOS technology show a high potential. CMOS technology allow the integration of the sensor together with control and image processing on the same chip, thus enabling the design of sensor systems at low cost. But modern sub-100nm technologies show also substantial disadvantages, such as increased leakage currents. Special circuit designs have been developed to especially reduce the influence of leakage currents. For application of the color correction method, new multi-channel photo sensors using vertically stacked photo diodes have been developed. The work further shows different concepts of multi-channel sensors capable of high quality color rendition. This approach is demonstrated on several new CMOS sensor designs with examples, implemented in a 90nm Infineon technology.

Color correction and grading are processes when fixing colors in recorded footage in Post-Production. The process of the two mentioned is a mix of technical adjustments and creativity. Color correction adjusts the colors between the clips/scenes so they match and look as natural or unique as possible. Grading is about the process of enhancing the look of a footage to achieve a certain style, it is of a more creative nature.Today, color correcting and grading are performed using desktop applications. The process means that recorded material is sent to the colorist from the set where it is received and downloaded to the computer where the work is performed. When the processing is considered complete, it is sent back to the recording team. This could be considered time consuming, and this process could be improved by using Accurate Video which is a web based program. Today, there are no features for color correction and grading in Accurate Video. The aim of this study was to design a User Interface (UI) for color correction and grading tools for Accurate Video application that meet the goals and needs of the people in this field of work, i.e. colorists. Based on literature studies including design guidelines, studies of what existing professional editing programs look like and what Accurate Video looks like, as well as interviews with colorists, a prototype was developed.

Color vision deficiency (CVD) is the inability or limited ability to recognize colors and discriminate between them. A person with this condition perceives a narrower range of colors compared to a person with a normal color vision. A growing number of researchers are striving to improve the quality of life for CVD patients. Finding cure, making rectification equipment, providing simulation tools and applying color transformation methods are among the efforts being made by researchers in this field. In this study we concentrate on recoloring digital images in such a way that users with CVD, especially dichromats, perceive more details from the recolored images compared to the original image. The main focus is to give the CVD user a chance to find information within the picture which they could not perceive before. However, this transformed image might look strange or unnatural to users with normal color vision. During this color transformation process, the goal is to keep the overall contrast of the image constant while adjusting the colors that might cause confusion for the CVD user. First, each pixel in the RGB-image is converted to HSV color space in order to be able to control hue, saturation and intensity for each pixel and then safe and problematic hue ranges need to be found. The method for recognizing these ranges was inspired by a condition called “unilateral dichromacy” in which the patient has normal color vision in one eye and dichromacy in another. A special grid-like color card is designed, having constant saturation and intensity over the entire image, while the hue smoothly changes from one block to another to cover the entire hue range. The next step is to simulate the way this color card is perceived by a dichromatic user and finally to find the colors that are perceived identically from two images and the ones that differ too much. This part makes our method highly customizable and we can apply it to other types of CVD, even personalize it for the color vision of a specific observer. The resulting problematic colors need to be dealt with by shifting the hue or saturation based on some pre-defined rules. The results for the method have been evaluated both objectively and subjectively. First, we simulated a set of images as they would be perceived by a dichromat and compared them with simulated view of our transformed images. The results clearly show that our recolored images can eliminate a lot of confusion from user and convey more details. Moreover, an online questionnaire was created and 39 users with CVD confirmed that the transformed images allow them to perceive more information compared to the original images.

<p dir="ltr">Blood hemoglobin (Hgb) testing is a widely used clinical laboratory test for a variety of patient care needs. However, conventional blood Hgb measurements involve invasive blood sampling, exposing patients to potential risks and complications from needle pricks and iatrogenic blood loss. Although noninvasive blood Hgb quantification methods are under development, they still pose challenges in achieving performance comparable to clinical laboratory blood Hgb test results (i.e., gold standard). In particular, optical spectroscopy can provide reliable blood Hgb tests, but its practical utilizations in diagnostics are limited by bulky optical components, high costs, and extended data acquisition time. Mobile health (mHealth) or diagnostic colorimetric applications have a potential for point-of-care blood Hgb testing. However, achieving color accuracy for diagnostic applications is a complex matter, affected by device models, light conditions, and image file formats.</p><p dir="ltr">To address these limitations, we propose biophysics-based machine learning algorithms that combine hyperspectral learning and spectroscopic gamut-informed learning for accurate and precise mHealth blood Hgb assessments in a noninvasive manner. This method utilizes single-shot photographs of peripheral tissue acquired by onboard smartphone cameras. The palpebral conjunctiva (i.e., inner eyelid) serves as an ideal peripheral tissue site, owing to its easy accessibility, relatively uniform microvasculature, and absence of skin pigmentation (i.e., melanocytes). First, hyperspectral learning enables a mapping from red-green-blue (RGB) values of a digital camera into detailed hyperspectral information: an inverse mapping from a sparse space (tristimulus color values) to a dense space (multiple wavelengths). Hyperspectral learning employs a statistical learning framework to reconstruct a high-resolution spectrum from a digital photo of the palpebral conjunctiva, eliminating the need for complex and costly optical instrumentation. Second, comprehensive spectroscopic analyses of peripheral tissue are used to establish a unique blood Hgb gamut and design a diagnostic color reference chart highly sensitive to blood Hgb and peripheral perfusion. Informed by the domain knowledge of tissue optics and machine vision, the Hgb gamut-based learning algorithm offers device/light/format-agnostic color recovery of the palpebral conjunctiva, outperforming the existing color correction methods.</p><p dir="ltr">This mHealth blood Hgb prediction method exhibits comparable accuracy and precision to capillary blood sampling tests (e.g., finger prick) over a wide range of blood Hgb values, ensuring its reliability, consistency, and reproducibility. Importantly, by employing only a digital photograph with the Hgb gamut-learned color recovery, hyperspectral learning-based blood Hgb assessments allow noninvasive, continuous, and real-time reading of blood Hgb levels in resource-limited and at-home settings. Furthermore, our biophysics-based machine learning approaches for digital health applications can lay the foundation for the future of personalized medicine and facilitate the tempo of clinical translation, empowering individuals and frontline healthcare workers.</p>