The relation between perceived image quality and naturalness was investigated by varying the colorfulness of natural images at various lightness levels. At each lightness level, subjects assessed perceived colorfulness, naturalness, and quality as a function of average saturation by means of direct category scaling. Colorfulness was found to increase monotonically with average saturation. The relation between the quality/naturalness judgments and average saturation could always be described by an inverted U-shaped function. A systematic difference was found between quality and naturalness judgments. This difference, reflecting the subjects' preference for more colorful, but, at the same time, somewhat unnatural images, was most noticeable at the original lightness level and diminished with decreasing lightness, in particular being least at the lowest lightness level investigated.

Image quality from electronic endoscopes is poor compared with that from conventional optical endoscopes that use color film. Our goal is to improve the image quality, particularly the color reproduction of electronic endoscopes. To achieve our goal, we have developed the endoscopic spectrophotometer to measure the spectral reflectance of gatric mucous membranes. Three hundred and ten spectra have been analyzed by principal component analysis. The results indicate that the reflectance spectra can be adequately described using only three principal components. Based on the above experimental result, we show that the spectral reflectance of gastric mucous membranes can be calculated from the R, G, B signals of conventional electronic endoscopes. Therefore, spectral reflectance of all pixels in gastric mucous membranes taken by electronic endoscopes can be estimated within less than the average color difference ΔE_uv^* = 2.66 in the L*u*v* color space. Computer simulations of the color reproduction of electronic endoscopes were performed, and the resulting spectral characteristics under various illuminants are described and analyzed.

New methods for both color palette design and dithering based on human visual system (HVS) characteristics are proposed. Color quantization for palette design uses the relative visual sensitivity and spatial masking effect of HVS. The dithering operation for printing uses nonlinear quantization, which considers the overlapping phenomena among neighbor printing dots, and then a modified dot-diffusion algorithm is followed to compensate the degradation produced in the quantization process. The proposed techniques can produce high-quality images in low-bit color devices.

Following a discussion of the disadvantages of today's color analysis in an open system architecture, a fundamentally improved reproduction system is proposed. This system includes multispectral sampling and multispectral color information transport via the communication net, using an encoded multispectral format compatible with conventional tristimulus color values. Further answers are given to the problems of the design of a multispectral sampling device. We show that only 14 normal interference filters are sufficient to analyze the spectra precisely enough. A multispectral encoding method produces not only a compatible data format to the conventional tristimulus color values, but is more effective than conventional encoding methods.

A new analytical method represents the surface of a color gamut for a reproduction process based on three or four primary colors by a single, closed expression directly in CIELAB. The method is based on the similarity of a color gamut to a cube (the CMY cube), and this similarity is represented by a kernel gamut. The kernel gamut is distorted by distortion and scaling functions in order to match the color gamut. The color gamut is fully represented by the distortion and scaling functions. The total number of their coefficients is below 150 at mean visual errors of below 2.15 ΔE_ab units.

Current subjective methods for evaluation and comparison of image processing techniques are generally not sufficiently relevant and accurate, essentially because they are not correlated with human observation. We propose a method to measure and display color image differences derived from the basic characteristics of human visual perception. This method provides images of color differences. These images can be used to define a global relative measure to predict image quality.

Color image analysis is still limited by the very significant amount of data. Most real-world images are, of course, not monochrome, but full color. Three-dimensional imaging utilizes large data sets, demanding computer storage and speedy algorithms of critical importance. With this aim in view, without taking into account both spatial and color information, a reducing step cannot be efficient. Thus, a multiresolution process seems to be well adapted, allowing simpler and faster computations. We present here a multiresolution tool, the Gaussian pyramid, first introduced for gray-scale images. We discuss the way to construct the color pyramid in a gamma-corrected RGB space, where the color mixing is additive.

For exact color reproduction of surfaces under various illuminants it is necessary to provide information on the complete reflectance spectrum. This work deals with the efficient representation of reflectance spectra. The number of values needed to represent a reflectance spectrum is reduced by linear transform coding of the spectra and by taking only a small number of transform coefficients. The aim is to find the basis functions for the linear transform such that an error criterion based on the perceptual color error difference is minimized. Depending on the error criterion different optimization schemes are employed and evaluated. The shape of the resulting basis function is analyzed and processed into a measure of roughness. This measure is then taken into account in the optimization, which results in smoother basis functions that lead to only a small degradation in the optimal color error values.