This PDF file contains the editorial “Editorial (April 1996)” for JEI Vol. 5 Issue 02

This PDF file contains the editorial “Guest Editorial: Digital Document Imaging” for JEI Vol. 5 Issue 02

This PDF file contains the editorial “Guest Editorial: Multimedia Systems” for JEI Vol. 5 Issue 02

Many algorithms have been proposed for detecting video shot boundaries and classifying shot and shot transition types. Few published studies compare available algorithms, and those that do have looked at limited range of test material. This paper presents a comparison of several shot boundary detection and classification techniques and their variations including histograms, discrete cosine transform, motion vector, and block matching methods. The performance and ease of selecting good thresholds for these algorithms are evaluated based on a wide variety of video sequences with a good mix of transition types. Threshold selection requires a trade-off between recall and precision that must be guided by the target application.

Motion-compensated visual pattern image sequence coding (MCVPISC) is an improved motion-compensated version of the visual pattern image sequence coding (VPISC) encoder. It is a high-performance video codec that is easily implemented in software. The MCVPISC codec is intended for real-time desktop computer applications such as multimedia delivery and local area network and point-to-point (modem) video conferencing. We describe a version of MCVPISC that has achieved bit rates of 0.15 bits/pixel or better for MPEG-2 test sequences (source encoding reduces this bit rate by about half). The computational complexity of the monochrome and color versions of the encoder is less than three integer operations per pixel. The decoder is bounded between 0.016 and 0.125 logical and integer operations per pixel. A prototype video conferencing system based on the MCVPISC codec has been implemented. The system hardware is comprised of two multimedia personal computers, two camcorders, two frame grabbers, and an ethernet connection. The prototype system software has a simple structure and runs under DOS and Windows™ 3.1. It includes a user interface, a video I/O interface, an event driven network interface, and a free running or frame synchronous video codec that also acts as the controller for the video and network interfaces. We have demonstrated that this prototype can support up to five full duplex full-motion video conferencing sessions. Future work will concentrate on expanding the prototype to support synchronous audio, multiple hardware and software platforms, and network protocols.

The enormous bandwidth potential of optical fiber has resulted in a worldwide effort to develop high-speed ATM networks, also called broadband integrated services digital networks (BISDN). Many of the applications that ATM networks will support will have a strong video component to them. Hence, it is important to understand the behavior of video traffic as it travels through these networks. To that end, we develop the generalized histogram model (GHM) to characterize "highly correlated" traffic, such as motion JPEG or "smoothed" MPEG traffic over ATM networks end-to-end. Using our GHM model we show how to determine the loss rate at any node in an ATM network. We find that, for highly correlated video sources, increasing the buffer size beyond a certain region called the "cell region" only marginally decreases the probability of loss. This implies that large buffers cannot be used to control the loss for such sources. The analytical model provided in this paper can be used for admission control, and network dimensioning and design in ATM networks. We have validated our results using simulations of real traces of video sources.

A method of improving the quality of service of video services in an asynchronous transfer mode network under congestion periods is considered. In the event of cell loss carrying intraframe coded data, the decoder asks the encoder to refresh the codec by intraframe coding the next frame. To prevent the network from overloading, constraints are imposed on the maximum number of sources that refresh the whole picture simultaneously at the same video frame interval and the distance between two consecutive refreshments in each video connection. The rate of this refreshment depends on the network load and the number of sources sharing the network. It is shown that there is a minimum distance within two consecutive refreshments for each video connection where both network performance and perceived image quality are optimized.

Spatial scalability is a feature referring to image representation in different sizes and is particularly useful in image browsing, as well as in progressive image transmission applications. We propose a new technique for scaling images based on combined wavelet transform and vector quantization (VQ). The proposed technique exploits the spatial scalability feature of wavelet transform and the efficient compression performance of vector quantization. Although VQ is a powerful technique for low bit rate image compression, the label stream VQ is not scalable. Hence, we first propose a scalable VQ algorithm (SVQ) where a smaller-size, coarse quality image can be obtained by decoding a portion of the label bitstream. This image can be further enhanced in size by progressively decoding the remaining bits of the label bitstream. We note that to ensure partial decodability of VQ labels, a multiresolution codebook structure is required. We then apply the SVQ technique in the wavelet domain (WSVQ) where the input image is wavelet-decomposed into three levels and the resulting coefficients are organized into vectors to exploit the intra- and interband correlations. The interband correlations are exploited by employing a nonlinear interpolative vector quantization. Simulation results confirm the superior subjective performance of the proposed technique at a significant reduced computational complexity.

The organization and management of multimedia information has been the subject of extensive research and development since the mid-1980s. However, a number of issues need to be addressed prior to developing a general purpose multimedia information system. We discuss an abstract architecture of such a system and highlight the issues of data modeling and multimedia information authoring, and also provide a systematic argument in favor of using an object-oriented approach over the relational approach as the basic framework for developing a multimedia information system.

Scanline algorithms are popular in computer graphics for complex geometric manipulations. The main characteristic of scanline algorithms is that a geometric transformation is decomposed into multipass transforms with each pass operating only along row or column scanlines. This leads to conversion of 2-D image manipulation problems to straightforward 1-D problems resulting in simple and systematic methods. The goal of this work is to examine the scanline approach for manipulation of transform-compressed images without decompressing them. We show how the scanline algorithms for rotation and projective mapping can be developed for JPEG/DCT images. The performance of the proposed scanline algorithms is evaluated with respect to quality, speed, and control and memory overhead.

A progressive region-based image compression technique is presented. This technique is based on the second-generation image compression paradigm, representing images as a union of homogeneous regions in harmony with the characteristics of the human visual system. The technique is progressive, allowing for an initial coarse version of the image to be reconstructed quickly from the compressed image data, and then refined through to the desired degree of fidelity. This paper describes how this technique can be applied to human face image archiving and retrieval. Compression results are provided and comparison with the JPEG standard is undertaken.

A morphological bandpass filter would, ideally, strictly limit the sizes of all features in an image to lie between the sizes of two similarly shaped but differently scaled structuring elements. A morphological bandpass decomposition of an image would be a disjoint set of morphological bandpass images with features of increasing size such that the set sums to the original image. Such strict bandpass limitations in size are not possible in general for arbitrary structuring element families. Hence, a true bandpass decomposition is not generally possible. Pseudo bandpass decompositions, in which intraband size limitations are relaxed, are possible and can be useful image analysis tools. Four pseudo bandpass image decompositions are described, one of which, the opening spectrum, is relatively well known and three of which are new. They are a decomposition derived from iteration of the top-hat transform, a morphological reconstruction of a Euclidean (quasi) granulometry, and a reconstruction of the opening spectrum. Properties of the opening spectrum and the top-hat transform are reviewed. The top-hat spectrum is defined, some of its properties are deduced, and it is compared to the opening spectrum. The reconstruction-based decompositions are defined and compared to the others. Comparative examples are given and a practical use described.

A multi-staged technique for locating the courtesy amount block on bank checks is presented. In the case of a check processing system, many of the proposed methods are not acceptable, due to the the presence of many fonts and text sizes, as well as the short length of many text strings. Particular methods chosen to implement a courtesy amount block locator (CABL) are described. First, the connected components in the image are identified. Next, strings are constructed on the basis of proximity and horizontal alignment of characters. Finally, a set of rules and heuristics are applied to these strings to choose the correct one. The chosen string is only reported if it passes a verification test, which includes an attempt to recognize the currency sign.

Optical character recognition (OCR) algorithms typically start from a binary label image. The need for a binary image is complicated by the fact that most imaging devices usually produce multiply valued data: a grey scale image. The problem then becomes how to extract the meaningful character data from the grey scale image. Image artifacts such as dirt, variations in background intensity, and imaging noise complicate the character extraction. When inspecting packages moving on a conveyor belt, we have control over the optical parameters of the system. Via autofocus and controlled lighting, parameters such as the optical path length, field of view, and illumination intensity may be adjusted. However no control can be placed on labels. The label reading system is totally subject to the package sender’s whimsy. We describe the development of a recurrent neural network to segment grey scale label images into binary label images. To determine a pixel label, the neural network takes into account three sources of information: pixel intensities, correlations between neighboring labels, and edge gradients. These three sources of information are succinctly combined via the network’s energy function. By changing its label state to minimize the energy function, the network satisfies constraints imposed by the input image and the current label values. The network has no knowledge of shape. Information on what comprises a desirable shape is probably unwarranted at the earliest stage of image processing. Although significant image filtering could be performed by a network that knows what characters should look like, such knowledge is unavoidably font specific. Further there is the problem of teaching the network about shapes. The neural network does not need to be taught. Learning is typically extremely time consuming. To be mappable to analog hardware, it is desirable that the neural equations be deterministic. Two deterministic networks are developed and compared.

This PDF file contains the editorial “Book Rvw: Wavelets and Subband Coding," by Martin Vetterli and Jelena Kovacevic for JEI Vol. 5 Issue 02