In an electronic color imaging device, such as a digital camera, using a single charge coupled device or complementary metal–oxide–semiconductor sensor, color information is usually acquired in subsampled patterns of red, green and blue pixels. Full resolution color is subsequently generated from this subsampled image. This is popularly called color interpolation or color demosaicing. In this paper, we present a color interpolation algorithm using a method of fuzzy membership assignment along with the concept of smooth hue transition. The algorithm is adaptive in nature and produces superior quality full resolution color images compared to most of the popularly known color interpolation algorithms reported in the literature. We present the results of a comparison with some challenging subsampled images for color interpolation.

Digital Still Color Cameras sample the color spectrum using a monolithic array of color filters overlaid on a charge coupled device array such that each pixel samples only one color band. The resulting mosaic of color samples is processed to produce a high resolution color image such that the values of the color bands not sampled at a certain location are estimated from its neighbors. This process is often referred to as demosaicking. This paper introduces and compares a few commonly used demosaicking methods using error metrics like mean squared error in the RGB color space and perceived error in the CIELAB color space.

Superpositions of periodic dot screens are largely used in electronic imaging in the field of color printing. In such superpositions the interaction between the superposed layers may cause new structures to appear which did not exist in any of the original layers: macrostructures (also known as moire´ patterns) and microstructures (also known as rosettes). While macrostructures are not always generated in the superposition (cf. moire´ -free superpositions), microstructures exist practically in any superposition, except for the most trivial cases. In fact, even the macrostructures, whenever they occur, consist of variations in the microstructure of the superposition. In the present paper we investigate the microstructures that appear in the superposition of periodic structures and their properties. We also find the conditions on the superposed layers under which the microstructure of the superposition varies—or remains invariant—when individual layers in the superposition are laterally shifted with respect to each other.

Wavelet transforms are often calculated by using the Mallat algorithm. In this algorithm, a signal is decomposed by a cascade of filtering and downsampling operations. Computing time can be important but the filtering operations can be speeded up by using fast Fourier transform (FFT)-based convolutions. Since it is necessary to work in the Fourier domain when large filters are used, we present some results of Fourier-based optimization of the sampling operations. Acceleration can be obtained by expressing the samplings in the Fourier domain. The general equations of the down- and upsampling of digital multidimensional signals are given. It is shown that for special cases such as the separable scheme and Feauveau’s quincunx scheme, the samplings can be implemented in the Fourier domain. The performance of the implementations is determined by the number of multiplications involved in both FFTconvolution- based and Fourier-based algorithms. This comparison shows that the computational costs are reduced when the proposed implementation is used. The complexity of the algorithm is O(N log N). By using this Fourier-based method, the use of large filters or infinite impulse response filters in multiresolution analysis becomes manageable in terms of computation costs. Mesh simplification based on multiresolution ‘‘detail relevance’’ images illustrates an application of the implemenentation.

In this work we develop a method for assessing the information density and efficiency of hyperspectral imaging systems that have spectral bands of nonuniform width. Imaging system designs with spectral bands of nonuniform width can efficiently gather information about a scene by allocating bandwidth among the bands according to their information content. The information efficiency is the ratio of information density to data density and is a function of the scene’s spectral radiance, hyperspectral system design, and signal-to-noise ratio. The assessment can be used to produce an efficient system design. For example, one approach to determining the number and width of the spectral bands for an informationefficient design is to begin with a design that has a single band and then to iteratively divide a band into two bands until no further division improves the system’s efficiency. Two experiments illustrate this approach, one using a simple mathematical model for the scene spectral-radiance autocorrelation function and the other using the deterministic spectral-radiance autocorrelation function of a hyperspectral image from NASA’s Advanced Solid-State Array Spectroradiometer. The approach could be used either to determine a fixed system design or to dynamically control a system with variable-width spectral bands (e.g., using on-board processing in a satellite system).

To evaluate the quality of images, most methods compare a degraded image to a perfect reference. Nevertheless in many cases, a reference does not exist. We propose an original univariant (i.e., without a reference) method based on the use of artificial neural networks. The principle behind it is to first teach a neural network to assess image quality using images taken from a pool of known examples, then use it to assess the quality of unknown images. The defects considered are compression artifacts, ringing, local singularities, etc. To simplify, only images with defects that are not mixed with one another were first used. Two illustrative examples are presented: assessment of the quality of JPEG compressed images and detection of local defects. The quality of the images is assessed without a reference and with error less than 6%–7% compared to the bivariant method that was learned. Our method can even be used to model some very simple visual comportment.

The increasing use of video compression standards in broadcasting television systems has required, in recent years, the development of video quality measurements that take into account artifacts specifically caused by digital compression techniques. In this paper we present a methodology for the objective quality assessment of MPEG video streams by using circular backpropagation feedforward neural networks. Mapping neural networks can render nonlinear relationships between objective features and subjective judgments, thus avoiding any simplifying assumption on the complexity of the model. The neural network processes an instantaneous set of input values, and yields an associated estimate of perceived quality. Therefore, the neural-network approach turns objective quality assessment into adaptive modeling of subjective perception. The objective features used for the estimate are chosen according to the assessed relevance to perceived quality and are continuously extracted in real time from compressed video streams. The overall system mimics perception but does not require any analytical model of the underlying physical phenomenon. The capability to process compressed video streams represents an important advantage over existing approaches, like avoiding the streamdecoding process greatly enhances real-time performance. Experimental results confirm that the system provides satisfactory, continuous-time approximations for actual scoring curves concerning real test videos.

Moving picture experts group (MPEG)-4 visual conformance standard specifies methods to verify whether bitstreams and decoders meet the requirements at a specified profile and level. The test of decoders can be divided into two parts: static and dynamic tests. The static test can be performed by comparing images from a decoder under test with those from a reference decoder using various test bitstreams. This paper proposes design methodologies of test bitstreams for a simple profile MPEG-4 texture decoder, and shows experimental results on the discrete cosine transform (DCT) scan type, DC/AC coefficient prediction, inverse quantization, inverse DCT, and various macroblock-type verifications.

This paper presents a novel approach to human gait analysis using a marker-free system. The devised acquisition system is composed of three synchronized and calibrated charge coupled device cameras. The aim of this work is to recognize in gray level image sequences the leg of a walking human and to reconstruct it in the three-dimensional space. An articulated threedimensional (3D) model of the human body, based on the use of tapered superquadric curves, is first introduced. A motion-based segmentation, using morphological operators, is then applied to the image sequences in order to extract the boundaries of the leg in motion. A reconstruction process, based on the use of a least median of squares regression is next performed, to determine the location of the human body in the 3D space. Finally, a spatial coherence is imposed on the reconstructed curves in order to better fit the anatomy of the leg and to take into account the articulated model. Each stage of the proposed methodology has been tested both on synthetic images and on real world images of walking humans. The obtained results suggest that this approach is quite promising and should be useful in the study of the gait.

Moving object extraction plays a fundamental role in computer vision, pattern recognition, object-based video coding and indexing. However, the low-level visual homogeneity criteria (like color, texture, intensity, motion, and so on) for segmentation do not lead to regions that immediately correspond to meaningful objects from the human point of view, because a semantic video object can contain totally different gray levels, colors, textures, or even motion. In this paper, a novel moving object extraction algorithm is proposed by collaborative integration of the results of a semantic object generation algorithm and a temporal tracking technique. First, the semantic objects are generated by a seeded region aggregation procedure according to some perceptual visual models or by a human– computer interaction procedure. Second, the correspondence of these extracted semantic objects along the time axis is further exploited through a temporal tracking procedure.

In this paper we describe a time efficient approach for computing the (3,4) distance transform and a method of producing intuitively well-shaped nonsensitive skeletons. The need and usefulness of abstracting both skeletal and distance transform information have been demonstrated in various earlier work. However, the approach presented here is intended to overcome several weaknesses while possibly permitting real-time computation on low-cost single or multiprocessor systems for applications such as video processing. Specifically, an incremental improvement to Kwok’s thinning algorithm is presented which allows the distance transform to be computed during thinning using significantly fewer addition and comparison operations. Additionally, efficient techniques are given which then further process the resultant skeleton using the computed distance transform information as well as information gathered about the surrounding chain codes. These techniques efficiently remove various skeletal artifacts, leaving well-shaped graph representations annotated with distance transform values.

Digital watermarking technology has loudly proclaimed its arrival over the past several years. It is difficult to find any journal associated with signals, imaging, sound, or video that doesn’t have a handful of digital watermarking related articles. Having worked in this field for several years now, I know that there is still quite a bit of confusion among users and policy makers about what this technology can and can’t do. Because the field is rather new, confusion is present even among the researchers. Signs of this confusion are multiple meanings for technical terms and rampant re-invention of basic algorithms.