One major disadvantage of multispectral imaging is the amount of data that needs to be stored for each multispectral image. In the past, several authors have proposed compression schemes to overcome this disadvantage. Most of these papers dealt with compression either in the spectral or in the spatial realm only, although some authors proposed to divide the data into luminance and color components and to subsample the latter, resulting in a combined method. Unfortunately, using multispectral imaging the luminance component depends on the illuminant used during the reconstruction of the color. Using a narrowband illuminant every single wavelength could correspond to the luminance information. This paper proposes a method that overcomes this difficulty by using a more basic approach. The reason, why it is possible to subsample the color channels in conventional color imaging is the dependence of the MTF of the human eye on the wavelength of the observed light. Therefore, it is possible to subsample each spectral band separately (yielding spatial compression). Furthermore, a spectral compression method is proposed, that is compatible to the spatial compression, resulting in a method that combines spatial and spectral compression of the multispectral image. Finally, the compatibility to conventional imaging will be addressed and it will be shown, that the new method allows achieving limited compatibility.
The present paper describes the framework and results of experiments aiming at developing computational methods to predict perceived image quality and perceived attributes, based on digital image data. Subjective judgments of overall contrast, sharpness, lightness, colorfulness, and quality were collected for 126 images presented on the monitor screen. For the analysis, digital images were represented in CIELAB color space, which was used as an approximation of the perceptually uniform color space. We utilized a computational vision approach and showed that all global attributes, including quality, can be predicted based on the combination of features (image properties) that could be considered as relevant elements at different levels of mental representation. The dependence of the global attributes upon the same features explains their mutual correlations, a known empirical fact. The regression results look very promising but require further refinement in terms of the feature assessment.
Natural Vision Image Data File Format has been developed in order to realize spectral-based color reproduction. This paper introduces the profile format of Natural Vision Image Data File Format. It is based on ICC Profile Format. New Profiles, Tags and Tag Types, therefore, have been developed to cover the lacks of ICC Profile Format for spectral-based color reproduction. The design concepts of Natural Vision System are introduced with respect to the requirements. Input Profiles and Display Profiles are expanded in order to provide spectral characteristics to color reproduction system. ColorSpace Conversion Profiles is expanded in order to provide rendering illuminant and CMFs to Color Reproduction Intent Management System. Relationships between the expanded Profiles and image processing are also descried. Introduction Spectral-based color reproduction system has more flexibility and reliability rather than colorimetric-based one from the viewpoint of colorimetry. The establishment of CIE Technical Committee 8-07 multispectral imaging symbolizes the importance of spectral-based color reproduction. Ohsawa has summarized the concept of spectralbased color reproduction system. The basic architecture of Natural Vision Image Data File Format also has been introduced as the solution of spectral-based color reproduction system. This paper introduces the details of the profile component. Relationship between the profile format and image processing is also explained. Design Concepts of Natural Vision System The following three requirements are established for the design of Natural Vision system: 1) To handle spectral data (R-1) 2) To be acceptable arbitrary illuminants (R-2) 3) To be acceptable arbitrary Color Matching Functions (CMFs) (R-3) 4) To be compatible with current system (R-4) Color reproduction system that realizes the above requirements has the following benefits: a) R-1 overcomes the problems of metameric matching. b) R-2 makes possible illumination exchanges. c) R-3 makes possible optimizations for individual CMFs. d) R-4 gives flexibility to Natural Vision System. Based on the above idea, Natural Vision system is designed as shown in Fig.1. To satisfy R-1, capturing system has to provide spectral reflectance or radiance of the object. Color Reproduction Intent Management System (CRI management system) is in charge of determining color reproduction target. To satisfy R-2 and R-3, CRI Management System has to be acceptable arbitrary illuminants (i.e. Etarget) and CMFs (i.e. ctarget) given by ColorSpace Conversion Profiles. In other words, color reproduction target (i.e. Sestimated,target or Cestimated,target) has to be able to define with the arbitrary illuminants and CMFs. The number of dimension of stimulus value C should be arbitrary. To satisfy R-4, the profile format is based on ICC profile format, AVI or BMP is used as the image format, and WAV format is used as the audio format in Natural Vision Image Data File Format. Figure 2 illustrates the details of the capturing system. It consists of multi-band camera, spectral estimation module and colorimetric estimation module. Input Profiles has to provide spectral characteristics of multi-band camera. The multi-band structure is useful to heighten the accuracy of the spectral estimation. Figure 3 illustrates the details of the CRI management system. It consists of spectral targeting module, colorimetric targeting module. The spectral targeting module is able to accept arbitrary illuminations (i.e. Etarget). The colorimetric targeting module is also able to accept arbitrary CMFs (i.e. ctarget). * The term “stimulus value” corresponds generalized (i.e. multidimensional) XYZ. Tristimulus value is a part of the stimulus value, where the number of dimension is three. ** The term “multi-band” means that the number of bands is arbitrary. Three-band camera (i.e. RGB camera) is included to the multi-band camera. It does not restrict that the number of the bands is over three. IS&T's 2003 PICS Conference
As the first step in historical research it is very important to read historical documents. Historians like to peruse original documents directly, but as in practice this may be difficult conventional photographic systems are often used. Historical documents vary greatly in type and size, therefore the application of digital imaging systems promises to contribute to the development of historical research. In this article, we at first point out the basic requirements of a digital imaging system used for historical research. Then according to these requirements, we introduce an image processing technique to improve the legibility of historical documents by spatial frequency filters. Since the paper used in some historical documents is very thin, letters written on the reverse side of the paper can be observed from the front side. As a result, letters on both sides of the paper become mixed up when the documents are read. This is one reason for the legibility of the documents to be degraded. We separate these two kinds of letters having different frequency components by two kinds of spatial frequency filters, which are fundamentally analogous to the un-sharp mask filter technique. These filters have different responsibilities in the frequency component to determine whether the letters are on the front side or not. The experimental results showed that the letters on the front side were separated well, but further considerations and additional experiments are necessary to improve the legibility of the historical documents.
In this paper, we present a novel approach for identifying and removing cast shadow in a color image. The technique employs clustering and color normalization procedures without the usual assumption that the darkest region constitutes a shadow or requiring the camera to be linear. The input image is transformed to a feature space spanned by the R-,G-,Bcolors and the Mean Shift Algorithm used for clustering. The number of such clusters denotes the number of significant distinct color regions including the shadow in the image. Color normalization has the tendency to remove shadow if a linear color space exists but does not if the space is non-linear. Using normalized color and Euclidean distance measure constraint, pairs of closest clusters called shadow candidate pairs are formed. Any shadow candidate pair whose distance apart is greater than the constraint is discarded as an invalid pair and vice-versa. The darker of the valid pair is regarded as shadow and can then be extracted. Our technique is also able to recover the background that is partly in shadow and partly illuminated. Image restoration is done by a mapping process whereby pixels that are resident in the shadow cluster are mapped to the mean of the ones they are closest to. We present some results using real color images with shadows on singleand multi-color backgrounds captured with a Kodak Zoom 120DC camera.
A standard color space, sRGB, has been widely used for reproducing the accurate color in the internet system. However, this color is accurate only under the sRGB reference viewing environment. In practice, the customers viewing environments are not always standard conditions. It is known that the multispectral color reproduction can be used to obtain highly accurate colors which are independent of device and viewing conditions. However, since the recent color reproduction system is based on conventional RGB or CMYK imaging technology, it is a huge task to change resources and devices in the market to be those in the multispectral technology. In our previous research, the spectral turn technique was proposed to achieve the spectral based color reproduction in the E-commerce with a high compatibility to the sRGB system. However the haploscopic method, used to compare colors on monitor and an object, was not compatible with the exact viewing conditions in real life. In this paper we propose more practical spectral turn technique using the revised mixed chromatic adaptation model, S-LMS(2001) applied with simultaneous binocular (SMB) matching method for more correct color appearance matching between softcopy and hardcopy under the practical viewing conditions.
The Volterra and Wiener theories of non-linear systems provide techniques with which to evaluate the behaviour of non-linear systems. It may be shown that the First Order Wiener Kernel Transform may be thought of as being equivalent to the linear portion of the MTF of a system. JPEG compression is such a non-linear system and as such measurement of its MTF has provided significant challenges. The First Order Wiener Kernel Transform of JPEG Compression (Version 6b) is determined under a variety of conditions and compared with SFR curves previously produced using ISO 12233. It is shown that there is agreement between the measurement methods. It is suggested that evaluation of the Wiener Kernel transform provides results that more closely represent the large-scale pictorial frequency response of JPEG compression due to the integration of intra and inter sub-image block effects. Further, that the technique represents an advantage over those previously considered.
The requirements of a multispectral imaging system are illuminated from an industrial point of view. Multispectral imaging finds its applications in the four basic fields archiving, faithful color image reproduction, color image communication, and color measurements. Example applications usually cover two or more of these four basic fields. They show the real need for multispectral imaging systems and the potential to reduce several color related problems to a minimum. Certain aspects influence the design of a multispectral imaging system. It must be stable, fast, easy to use, and safe against user mistakes. Multispectral images must be selfcontained and need to be stored with varying levels of accuracy. Moreover, the lack of spectral output devices gets evident as the accuracy of the rendered multispectral images achieves a certain level.
3D Virtual Reality tool makes color analysis intuitive. Indeed color data are much easier to analyze using 3D diagrams than using 2D sections or inspecting directly the ASCII data files. The aim of this study is to propose a set of tools based on 3D Virtual Reality to display color information in differents color spaces and different color representations, first in order to analyze which color space is the most relevant in regards to a computer vision process and a given image, and next to test in realtime differents algorithms to select the most relevant algorithms, or to understand why an algorithm or a color space gives better results than another, in regards to a given process (segmentation, quantization, indexing, ...) and a given image.
In this paper, a spectral color appearance model is considered. This model is used for color image quality estimation when a color image is reproduced through a spectral image. The model is based on the statistical image model that sets a relationship between the parameters of the spectral images and color images, and the overall appearance of the image. This was used to create a set of test color images that were evaluated by twenty color normal test persons. The results of their evaluation were used for synthesis of a Fuzzy Logic System. Based on the test persons estimation two-dimensional colorfulness vividness quality evaluation surfaces were built for each image and for image groups. The study shows that a spectral-color relationship and Fuzzy Logic Inference are efficient in color image quality evaluation.
A constant exposure method for DSC IGBT strobe is proposed. This method is based on two lookup tables (exposure and energy table). In this method, light box is used to generate exposure table and different flash time is used to generate energy table. Through exposure table and preflash result, we can get digital number ratio of preflash to main flash. This ratio and energy table can be used for predicting the strobe main flash time under different brightness conditions.
In the image processing of a digital camera system, a colorimetric matching transformation is implemented to minimize the error between the sensor output transformed color responses and the human vision responses. The color matching transform does not usually attempt to minimize the resulting noise of the color corrected data. The CMOS sensors used in embedded applications typically exhibit high levels of random noise and cross-talk (the loss of photons or electrons from a pixel to neighboring pixels). This results in a trade-off between color error (e.g., desaturated colors) and the amount of noise in the image. This is especially true as the size of pixels are progressively reduced, which leads to highly de-saturated (muted) images that require severe color corrections and have lower signal-to-noise ratios. In this paper, methods of color correction are presented which exploit the fact that the human visual system is most sensitive to color errors in lower spatial frequencies than in higher spatial frequencies. Thus more accurate (and usually more complex) color correction algorithms should be applied to the lower spatial frequency data, and less accurate (and usually simpler) algorithms can be applied to the higher spatial frequency data. Three methods are developed and analyzed which use segmentation of sensor data in the spatial frequency domain: the discrete cosine transformation (DCT), the discrete wavelet transformation (DWT), and a simple low pass/high pass filter. Metrics for noise performance, color accuracy, and image sharpness are provided. The novel approach developed in this paper is the use of variable complexity color corrections applied within the frequency transform domains. This allows the process of compressing image data to be used to reduce the amplification of uncorrelated pixel noise, while still achieving accurate color enhancement and maintaining image sharpness.
Using a spectroradiometric model of capture for a digital camera based on the mathematical description of the empirical opto-electronic conversion spectral functions (OECSF), the capture of MacAdam or optimal spectra with fixed illumination level is simulated. This model of capture allows to change freely the f-number of the zoom-lens and/or the photosite integration time of the electronic shutter of the camera, regardless of the spectral composition of the stimulus. If we follow the procedure employed by MacAdam in 1935 working with the CIE1931 XYZ standard observer, these color-stimuli are arranged in decreasing pyramidal form as the luminance factor increases for any chromaticity diagram (CIE-xy, UCS-u'v' or CIE-L*a*b*). These loci are often called MacAdam limits or Rosch color solid. On the other hand, transforming the simulated RGB digital output levels of the optimal colors to XYZ data through the raw colorimetric profile with luminance adaptation of our digital image capture device, the corresponding MacAdam loci for each luminance factor are smaller than those of the colorimetric standard observer. This systematic desaturation of the optimal color-stimuli shows that our color device, in raw performance, desaturates in general the real color-stimuli, so this result justifies the additional use in digital photography of color correction algorithms, more or less complex, in order to reach the colorimetric status of color reproduction.
In this paper, it will be shown that the effect of color saturation and hue on image quality can be described in a very simple way. From measurements given in this paper it appears that there is a remarkable agreement between color saturation and gamma with respect to their effect on image quality. From an earlier investigation it was found that image quality increases with the square root of gamma up to an optimum value of gamma and decreases with the inverse of the square root of gamma at higher values. It now appears that the eye reacts in a similar way on an increase of colorfulness, so that both effects can be described by the same type of equations. From measurements on the effect of hue on image quality it appears that the image quality decreases linearly with the angular rotation of the color coordinates in CIELUV or CIELAB space, if the image quality is expressed in jnds. This decrease varies symmetric with the direction of the rotation. Introduction As described in a previous paper, the eye reacts in a nonlinear way on luminance variations in an image around the average luminance. This effect can be explained by the voltage response of the cones at a variation of luminance. An example of this response is given in Figure 1. This figure shows the voltage response of the cones of a turtle measured by Burkhardt for a single adaptation luminance. At a different adaptation luminance, the curve shifts to that level. As the cones of a turtle behave similarly as that of humans, these data can be used for a general description of the visual response of the eye at a variation of luminance. The measured voltage variation can be described by the following equation: ad L c L L V + = (1) where V is the voltage expressed in relative units varying from 0 to 1, L is the luminance, Lad is the adaptation luminance, and c is a constant close to 1. The exponential slope γ of this relation can be calculated as follows: ad 1 1 ) (ln ) (ln L c L dL dV V L L d V d + = = = γ (2) For L = cLad, V = 0.5 and γ = 0.5. This means that the voltage varies with the square root of the luminance around the average luminance of an observed image. This nonlinear behavior of the visual system was taken into account in the SQRI, or square-root integral, for the description of image quality. At the time of development of this method, the here given data were not yet available, but perceptual data indicated already this behavior. The SQRI is given by the following equation: ) d(ln ) ( ) ( 2 ln 1 max min t u u m u M J u
For serious imaging practitioners, the benefits of variety and economy in digital capture device selection come as a mixed blessing, among dizzying performance specification hype. This “specsmanship” has created a bazaar-like atmosphere where manufacturers’ claims of resolution, speed, and dynamic range resonate like those of so many market barkers. Through regulation, education, and enablement, sciencebased performance standards vetted through ISO/TC42 will allow many of these claims to be supported or refuted. This paper details the technical content of these efforts and the challenges required to make these standards rugged and easily implemented, as an aid toward less ambiguous and informed device selection.
In this study two measures for defining color differences in spectral space are defined using two spectral databases, Munsell Glossy and NCS. First of the measures is a Ndimensional Euclidean distance between two radiance spectra. Spectral differences of constant chroma, adjacent hues and adjacent values in Munsell and NCS-databases are evaluated and analyzed based on this measure. Three-dimensional conical color-space with first three PCA-eigenvectors of NCSand Munsell data as basis vectors is defined and analyzed. The second error measure is defined as the Euclidean distance in this space. Similarities between eigenvectors and opponent signals proposed by Hurvich and Jameson are noticed. Smoothed eigenvectors of NCS are concluded to be better for creation of a uniform color-space than the Munsell eigenvectors. The projections of a radiance spectrum to the modified eigenvectors define the coordinate values of the colorspace. It is noticed that by weighting the first and the second eigenvector by luminous efficiency curve the colorspace will be more uniform. Finally the variables are modified so that the color-space would be as uniform as possible but still allowing the calculations be quite simple. Also simple chromatic adaptation terms are defined for this color-space to improve its performance. Hue angle differences between adjacent hues of Munsell value 6 are determined using standard color-spaces and modified 3dimensional spectral eigenvector space. Also the chroma scales as a function of hue are evaluated. Chroma and hue differences in spectral space and in modified eigenvector space are compared to the most common color difference formulas (CIELAB E, CIE94 and CIEDE2000). Performance of defined color-space with Munsell Glossy spectra is compared to the CIELAB-space and the CIECAM97s-model. Introduction A majority of color difference formulas are based on the CIELAB-color-space. The original color difference formula in the CIELAB-space was defined as an euclidean distance in a 3D-space. Afterwards different kind of terms and parameters have been added to improve the performance of the color difference formula. 2, 3 Recently, many kinds of color appearance models have been constructed. These include ATD-models and CIECAM97s-model, which is recommendation of CIE. Although one claims, that CIECAM97s is uniform, it still hasn’t been used much for color difference definitions. One reason may be its complexity. Furthermore, CIECAM97smodel has been revised many times since the original version was published, and a completely new version, CIECAM02s, has been published recently. Also a new color difference formula based on CIELAB, CIEDE2000, was developed. CIEDE2000-color difference formula is also recommended by CIE. The color difference formulas based on CIELAB cannot be explained physiologically. They are formulas, which improve mathematically the uniformity of the CIELAB-space. However, we can’t define any logical coordinate system with those formulas. An ideal uniform colorspace should base on the physiology of human visual system. The lack of information about the human visual system makes it impossible to create an unarguable color appearance model on physiological basis. The second chance to create uniform color-space is to use statistical methods. We can create three-dimensional dataset by calculating the first three principal components for radiance spectra of the dataset. Recently this kind of PCA-based spectral analysis has been done much in the field of spectral imaging. By editing these eigenvectors we are able to define almost the same dimensions as the human color vision system has. Here one needs to point out that usually our databases are based on some physiological theory or model, for example in case of Munsell system that is the opponent theory model. It is well known that the first eigenvector of spectral dataset is proportional to the mean of the spectra and further to the intensities of the spectra. The second and the third eigenvectors are quite similar to the opponent signals proposed by Hurvich and Jameson. Another way to create uniform colorspace statistically is to use optimizing methods which map spectra to the uniform color space defined by uniform dataset. Experimental The coordinate systems were defined using two spectral datasets, Munsell Glossy data (1600 samples) and NCSdata. In case of Munsell Glossy dataset, reflectance spectra of colorpatches were measured by a spectrophotometer from wavelength 380 to 780 nm with 1 nm intervals. Radiance spectra were calculated using C-illumination, to which Munsell data is calibrated. In case of NCS data, IS&T's 2003 PICS Conference
Focus is an important part before capturing an image by DSC. After obtaining an optimized focus, image can have best contrast to evaluate. So far, many focus methods were built and they include maximum contrast search and object distance inverse calculation. No mater which method is used, both can provide us best focus to capture. In this paper, another method to determine focus is moiré. Reason of moiré happened is related to sensor’s pixel pitch and spatial frequency of capture target. When moiré happened and based on mosaic CFA (Color Filter Array), the captured image will result in special slanted line and some colored pixel. In some focus test, moiré was known to be a method to determine focus, but it is inspected by human visual without any quantification. This paper tries to use chroma of moiré to make some experiment and to compare with contrast method. Therefore, focus can be determined simply by chroma calculation
Modern digital imaging techniques have been applied to the recovery of erased and overwritten writings on historical parchment manuscripts. In our work, we have found that images obtained under different illuminations can increase the contrast of erased or faded writings on parchment. This paper focuses on the use of ultraviolet illumination, which causes parchment to fluoresce, emitting visible light towards the blue end of the spectrum. This fluorescence increases the contrast of both faded and erased iron gall inks on parchment. On the thousand-year-old Archimedes Palimpsest, fluorescence imaging, together with post-capture image processing, makes much of the erased Archimedes text visible to the classical mathematics scholars who are trying to read it. In addition, we have found that ultraviolet reflectance imaging has been able to reveal characters on a scraped page of a five-hundred-year-old Hebrew prayer book that showed little evidence of characters under ultraviolet fluorescent imaging. The physical mechanisms for these differences are not well understood, but they can be used to recover writings that otherwise would be lost forever. Introduction There are many old manuscripts in libraries and private collections that have been damaged, erased, or have just faded with time. Scholars would very much like to read these writings, which in some cases may be the only remaining copy of an ancient author’s work. A traditional method to read faded writing – one that scholars have used for Keith T. Knox and Roger L. Easton, Jr. Chester F. Carlson Center for Imaging Science, Rochester Institute of Technology, Rochester, New York
Artists have different parameters to, for example, colour scientists, when considering the quality of the finished image. An analysis of the work is based on an artist's conception of the work and a subjective assessment of print surface, colour and image quality, which although may appear to be based on the same criteria as a reproduction, the impact on the viewer is quite different. This presentation uses The International Digital Miniature Print Portfolio as a case study, which illustrates issues relating to: implications of printing digital files from unknown sources; generating, saving of images from unknown sources; file tagging; converting unrecognisable generic colour profiles; bespoke colour profiles for artists' handmade papers and mouldmade papers; compatibility of paper and ink. As a result of working on the portfolio, a method for optimising the print workflow has included: lightfast testing on a variety of papers; bespoke ICC profiles for particular papers; a best practice for artists to generate, save and print images.
