Edward R. Dougherty Editor

The color-image quality of color overhead transparencies depends on properties of the imaging system used to create the transparency and illuminating and viewing conditions of the transparency such as the projector’s spectral power distribution, projector distance from the screen, and luminance, ambient lighting, screen gonio-spectral reflectance factor, and viewing distance and geometry. As different visual fields and/or luminance of those fields, some of these illuminating and viewing conditions can be taken into suitable account using color-appearance models. A visual experiment is performed to determine whether color-appearance correlates of visual perception could be used to predict color-image quality for this imaging modality. The Hunt 1991 color-appearance model is used to define correlates of hue, brightness, colorfulness, lightness, and chroma for both pictorial and business-graphic scenes viewed under several combinations of ambient illuminance and projector luminance. Gamut volume is defined based on either absolute attributes—hue, brightness, and colorfulness—or relative attributes—hue, lightness, and chroma. Seventeen observers performed a preference experiment generating interval scales of color-image quality. It is found that gamut volume defined by using correlates of hue, brightness, and colorfulness well predicted color-image quality. Of these correlates, colorfulness was the most important factor.

Two error diffusion algorithms, based on Pappas's printer model accounting for dot-overlapping and ink distortion, are presented to achieve good color reproduction. The basic idea is to combine printer and color models on the perceptually uniform Commission Internationale de l'Eclairge (CIE) L*a*b* (CIE 1976) color space. The models, derived from the Neugebauer equations and color matching theories, are designed to achieve the minimization of the human visual color distortions between the colors of original pixels and those of a halftoned image. The effectiveness of our approaches is shown by comparison and examination of two error diffusion algorithms with previous methods: the error diffusion based models and the window based minimization algorithm. Experimental results of the error clipping technique, focused on the real application of the nonseparable algorithm, and the desired range of error clipping, where an image produced by the nonseparable algorithm can be stable without additional color distortion, are reported.

We present a novel nonlinear predictive image coding scheme in which a relative prediction error is first generated from the current pixel value and its predicted value. It is next mapped, quantized, coded and transmitted. Consequently, a weighting function is introduced into the coding algorithm such that the coding error is adapted by the pixel intensity and its relative prediction error. Meanwhile, the resulting quantization step size is smaller in lower contrast areas and larger in higher contrast areas so that the granular noise and the slope overload distortion can be efficiently reduced. Our simulation results show that on an average, with the proposed scheme, the bit rate is about 0.23 bits less than that obtained with differential pulse-code modulation (DPCM), while the peak SNR (PSNR) is about 2.9 dB higher than that with DPCM. On the other hand, more coding errors are allocated in less visible areas where the image intensity and/or contrast are higher.

A companding procedure designed to improve the visual quality of monochrome images encoded with the baseline Joint Photographic Experts Group (JPEG) image compression algorithm is described. The companding procedure consists of two pointwise nonlinearities applied to the image, one before and one after the JPEG encoding. In a series of two-alternative forced-choice experiments designed to measure human visual threshold response to these images, it was determined that the average bitrate at the just-noticeable- difference (JND) point for the companded images is less than that for the uncompanded images. In a suprathreshold experiment, the subjects selected the companded images as being less distorted than the uncompanded images in more than 80% of the trials, for image bitrates ranging from 0.4 to 1.0 bpp.

Lossless compression techniques are essential in some applications, such as archival and communication of medical images. In this paper, an improvement on the Joint Bi-Level Imaging Group (JBIG) method for continuous-tone image compression is proposed. The method is an innovative combination of multiple decorrelation procedures, namely a lossless Joint Photographic Experts Group (JPEG)-based predictor, a transform-based inter-bit-plane decorrelator, and a JBIG-based intra-bit-plane decorrelator. The improved JBIG coding scheme outperformed lossless JPEG coding, JBIG coding, and the best mode of compression with reversible embedded wavelets (CREW) coding, on the average bit rate, by 0.56 (8 bits/component images only), 0.14, and 0.12 bits per pixel with the JPEG standard set of 23 continuous-tone test images. The compression technique may be easily incorporated into currently existing JBIG-based products. A high-order entropy estimation algorithm is also presented, which indicates the potentially achievable lower bound bit rate, and should be useful in decorrelation analysis as well as in the design as cascaded decorrelators.

Blue-noise dither halftoning methods have been found to produce images with pleasing visual characteristics. Results similar to those generated with error-diffusion algorithms can be obtained using an image processing algorithm that is computationally much simpler to implement. The various techniques that have been used to design blue-noise dither matrices are reviewed and compared. In particular, a series of visual cost function based methods and several techniques that involve designing the dither matrices by analyzing the spatial dot distribution are discussed. Ways to extend the basic blue-noise dither techniques to multilevel and color output devices are also described, including recent advances in the design of jointly optimized color blue-noise dither matrices.

A public domain optical character recognition (OCR) system has been developed by the National Institute of Standards and Technology (NIST). This standard reference form-based handprint recognition system is designed to provide a baseline of performance on an open application. The system’s source code, training data, performance assessment tools, and type of forms processed are all publicly available. The system is modular, allowing for system component testing and comparisons, and it can be used to validate training and testing sets in an end-to-end application. The system's source code is written in C and will run on virtually any UNIX-based computer. The presented functional components of the system are divided into three levels of processing: (1) form-level processing includes the tasks of form registration and form removal; (2) field-level processing includes the tasks of field isolation, line trajectory reconstruction, and field segmentation; and (3) character-level processing includes character normalization, feature extraction, character classification, and dictionary-based postprocessing. The system contains a number of significant contributions to OCR technology, including an optimized probabilistic neural network (PNN) classifier that operates a factor of 20 times faster than traditional software implementations of the algorithm. Provided in the system are a host of data structures and low-level utilities for computing spatial histograms, least-squares fitting, spatial zooming, connected components, Karhunen Loève feature extraction, optimized PNN classification, and dynamic string alignment. Any portion of this standard reference OCR system can be used in commercial products without restrictions.

An integrated program development environment for computer vision tasks is presented. The first component of the system is concerned with the visualization of 2-D image data. This is done in an object-oriented manner. Programming of the visualization process is achieved by arranging the representations of iconic data in an interactively customizable hierarchy that establishes an intuitive flow of messages between data representations seen as objects. The visualization objects, called displays, are designed for different levels of abstraction, starting from direct iconic representation down to numerical features, depending on the information needed. Two types of messages are passed between these displays (update and result messages), which yield a clear and intuitive semantics. The second component of the system is an interactive tool for rapid program development. It helps the user in selecting appropriate operators in many ways. For example, the system provides context sensitive selection of possible alternative operators as well as suitable successors and required predecessors. For the task of choosing appropriate parameters several alternatives exist. For example, the system provides default values as well as lists of useful values for all parameters of each operator. To achieve this, a knowledge base containing facts about the operators and their parameters is used. Second, through the tight coupling of the two system components, parameters can be determined quickly by data exploration within the visualization component.

Joan L. Mitchell, William B. Pennebaker, Chad E. Fogg, and Didier J. LeGall. 512 pages. ISBN 0-412-08771-5. Chapman and Hall, New York ~1997! $69.95 hardbound. Reviewed by Glen G. Langdon, Univ. of California, Santa Cruz, Computer Engineering Department, Applied Science Building, Room 225, Santa Cruz, California 95064-1078