This paper presents a study investigating the reconstruction of cross sectional images of a time-varying object from its projections. Ideally, to obtain high quality cross sectional images of a time-varying object, the projections measured should correspond to that of the same phase of the object. This in turn requires that the projections from different projection angles be collected at a high rate—a requirement that is difficult to achieve for various practical reasons. In this paper, we use a priori knowledge of underlying object and the spectral properties of the projections to overcome the deficiency of inconsistent projections, and develop a novel image reconstruction method. Computer simulations demonstrate that the proposed method is capable of reducing the reconstruction mean squared error and generating reconstructions with much improved quality.

This paper describes a video analysis system, free of markers and setup procedures, that quantitatively identified gait abnormalities in real time from standard video images. A novel color three-dimensional body model was sized and texture mapped to the exact characteristics of a person from video images. The kinematics of the body model was represented by a transformation tree to track the position and orientation of a person relative to the camera. Joint angles were used to track the location and orientation of each body part, with the range of joint angles being constrained by associating degrees of freedom with each joint. To stabilize tracking, the joint angles were estimated for the next frame. The calculation of joint angles, for the next frame, was cast as an estimation problem, which was solved using an iterated extended Kalman filter. Patients with dopa-responsive Parkinsonism, and age-matched normals, were video taped during several gait cycles with walking movements successfully tracked and classified. The results suggested that this approach has the potential to guide clinicians on the relative sensitivity of specific postural/gait features in diagnosis.

Some large field digital radiography systems are currently under development using multiple detectors. These small size two dimension detectors are abutted together to cover a large field. Physical gaps existing between adjacent detectors produce seams between the resultant subimages. In this paper, an adaptive linear interpolation algorithm was introduced for estimating missing information at the seams and was compared with conventional linear interpolation and with nonlinear interpolation. The effectiveness of the algorithms was evaluated for relative/absolute errors. Phantom images were acquired using prototype digital radiography systems. Seams with width ranging from two pixel to six pixel were introduced and adaptive linear interpolation algorithms were applied to estimate missing information at the seams. Quantitatively, the adaptive interpolation offers at least equivalent or less error than that of the linear interpolation algorithm. The experimental results prove that there is a significant difference between two algorithms for tested radiographic images. The comparison results also show that the adaptive interpolation offers better performance than nonlinear interpolation. When developing large field digital radiography imaging systems, gaps between adjacent detectors should be minimized. For narrow seams, the adaptive linear interpolation algorithm is a practical solution because of its simplicity and effectiveness.

Most face databases are meant for pattern recognition use. We report the creation of a unique face database which may be used for color-related studies of faces. We describe it as physicsbased because it not only contains color images but also data on color image formation such as spectral reflectance measurements from facial skin, illuminant spectral power distribution, and camera spectral response. We demonstrate the usefulness of the database by an approach we developed to color correct face images taken under different illumination conditions.

Texture synthesis is the ability to create ensembles of images of similar structures from sample textures that have been photographed. The method we employ for texture synthesis is based on histogram matching of images at multiple scales and orientations. This paper reports two fast and in one case simple algorithms for histogram matching. We show that the sort-matching and the optimal cumulative distribution function (CDF)-matching (OCM) algorithms provide high computational speed compared to that provided by the conventional approach. The sort-matching algorithm also provides exact histogram matching. Results of texture synthesis using either method show no subjective perceptual differences. The sort-matching algorithm is attractive because of its simplicity and speed, however as the size of the image increases, the OCM algorithm may be preferred for optimal computational speed.

In most images processed using error diffusion techniques, a nonhomogeneous arrangement of dots (‘‘worm’’ artifacts) may occur for highlight and shadow regions. In this article we propose a method to reduce the worm artifacts using spatial constraints for the dot arrangement. For a pixel in the shadow or highlight region, a dot is placed only if a minimum distance constraint between the current pixel and the placed dots is satisfied. If the distance constraint is not satisfied, the placement of the current dot is postponed and the quantizer error is diffused to the unprocessed pixels. The proposed method verifies the distance constraint without computing geometric distances. Instead, a roadmap enlarging from the current pixel location is used to check for placed dots. A dot is placed only if no other dot along the scanned roadmap is found. The scanned length of the roadmap depends on the input gray level and is accessed from a look-up table. The paper includes an example of halftone image processed with the proposed method.

The emerging international standard for compression of bilevel images and bilevel documents, JBIG2, provides a mode dedicated for lossy coding of halftones. The encoding procedure involves descreening of the bilevel image into gray scale, encoding of the gray-scale image, and construction of a halftone pattern dictionary. The decoder first decodes the gray-scale image. Then for each gray-scale pixel the decoder looks up the corresponding halftone pattern in the dictionary and places it in the reconstruction bitmap at the position corresponding to the gray-scale pixel. The coding method is inherently lossy and care must be taken to avoid introducing artifacts in the reconstructed image. We describe how to apply this coding method for halftones created by periodic ordered dithering, by clustered dot screening (offset printing), and by techniques which in effect dithers with blue noise, e.g., error diffusion. Besides descreening and construction of the dictionary, we address graceful degradation and artifact removal.

The two major aspects of image data compression utilizing wavelet analysis and synthesis are the decomposition of an image and the reconstruction of this image. It has been noticed in this investigation that the pyramid structure of convolution and the down sampling or the up sampling (adding zeros) and convolution have equivalent operations in vector space analysis. That is, the decomposition is equivalent to an outer product expansion. Therefore, tensor products can easily accomplish the synthesis or reconstruction. This is sometimes called the direct product. It is suggested that this method of implementation saves operations and opens the way to utilization of uniform filter banks.

In this paper, the captions for several color figures were inadvertently left out of the journal. Those figures are reproduced on the following two pages in black and white with the appropriate captions. Please refer to the original paper for color renderings of the figures (Vol. 8, No. 3, pp. 274–275). We regret this error.

Since Fourier analysis is now commonly applied to imaging systems, one might assume that there is no longer sufficient interest for a book devoted entirely to the optical transfer function (OTF).