The recent development of video compression algorithms allowed the diffusion of systems for the transmission of video sequences over data networks. However, the transmission over error prone mobile communication channels is yet an open issue. In this paper, a system developed for the real time transmission of H263 video coded sequences over TETRA mobile networks is presented. TETRA is an open digital trunked radio standard defined by the European Telecommunications Standardization Institute developed for professional mobile radio users, providing full integration of voice and data services. Experimental tests demonstrate that, in spite of the low frame rate allowed by the SW only implementation of the decoder and by the low channel rate a video compression technique such as that complying with the H263 standard, is still preferable to a simpler but less effective frame based compression system.

In web production a continuous work flow is very important, specially in paper-mills that produce high quality photo papers. Often rupturing of the paper-web cause falling-off for hours. Knowledge about the reasons for this kind of failure helps to tune the production parameters appropriately to avoid interrupts and enlarge the mtbf coherently.

This paper presents some approaches intended to maximize the available processor bandwidth of a real time video surveillance system. The techniques approach this goal from data centric and process centric perspectives. Data vectorization focuses on organizing and transformign data to more efficiently process it. Some cache considerations and compressed data analysis techniques are therefore reviewed. Dynamic scheduling focuses on using application specific information to reduce the iterations are complexity of repeated processes. Some novel applications of these techniques to video tracking gand recognition are presented. Some implementation examples are also provided indicating that the trade-offs in such an implementation are economically viable.

We propose EVLIW as a new processor architecture which is designed for general purpose processing and is suitable especially for real-time image processing. The processor architecture is a VLIW, but it has more functional units than the generic VLIW processor has. The EVLIW consists of the interconnection network for connecting the neighbor and of functional units, which are more primitive than in the generic VLIW processor. Some of general-purpose processors in the market includes several processing units, e.g. the same four single precision floating-point or four 16bit-word integer units for Intel processor with SSE/MMX, where the four units do the same operation with the four different data. In the image processing, the data are processed in parallel, where the operating is not complicated an only the high-speed processing is usually required. We have tried a simple image processing using Intel's processor with SSE/MMX and summarize the results. In this paper, we describe a new architecture for real-time imaging, and its design, comparing with Intel's processor with SSE/MMX.

In this article, we present a popular loseless compression/decompression algorithm, GZIP, and the study to implement it on a FPGA based architecture. The algorithm is loseless, and applied to 'bi-level' images of large size. It insures a minimum compression rate for the images we are considering. The proposed architecture for the compressor is based ona hash table and the decompressor is based on a parallel decoder of the Huffman codes.

In this paper, an FPGA based architecture for stereo vision is discussed. The architecture provides a high-density disparity map in real time. The architecture is based on area comparison between the image pair using the sum of absolute differences. The architecture scans the input images in partial columns which are then processed in parallel. The system performs monolithically on a pair images in real time. The purpose of the system is to be integrated ins smart camera for real-time image analysis based on FPGA processing.

In this paper, we describe a novel approach to image sequence segmentation and its real-time implementation. This approach uses the 3D structure tensor to produce a more robust frame difference signal and uses curve evolution to extract whole objects. Our algorithm is implemented on a standard PC running the Windows operating system with video capture from a USB camera that is a standard Windows video capture device. Using the Windows standard video I/O functionalities, our segmentation software is highly portable and easy to maintain and upgrade. In its current implementation on a Pentium 400, the system can perform segmentation at 5 frames/sec with a frame resolution of 160 by 120.

This paper presents a modification of the multi-scale clustering algorithm in order to reduce the color content of an image in an automatic manner without requiring the specification of the number of colors. The clustering space is chosen to be the YCbCr color space as used in the JPEG compression standard. Although multi-scale clustering is capable of defining prominent data clusters in an automatic manner, it may not generate all perceptibly distinctive colors when applied to the YCbCr color space. In this work, we have modified the multi-scale clustering algorithm to overcome this limitation for color reduction purposes. Our color reduction algorithm consists of two parts. The first part is a modification of multi-scale clustering to obtain a set of primary color prototypes. The second part is a color sectoring method to obtain a set of secondary color prototypes. The developed color reduction algorithm has been implemented on a high performance DSP processor, namely TMS320C6201, due to the computational requirement associated with multi-scale clustering. The result show on average 50 percent more compression over that of the JPEG standard for the color portion of images with comparable levels of color distortion.

Optimal translation-invariant binary windowed filters are determined by probabilities of the form, where x is a vector of observed values in the observation window and Y is the value in the image to be estimated by the filter. The optimal window filter is defined by y(x) equals 1 if P(Y equals 1/x) > 0.5 and y(x) equals 0 if P(Y equals 1/x) <EQ 0.5, which is the binary conditional expectation. The fundamental problem of filter design is to estimate P(Y equals 1/x) from data, where x ranges over all possible observation vectors in the window. A challenging aspect of optimal translation- invariant binary windowed filters is the implementation for large windows. In the context of Bayesian multiresolution filter design recently published by the authors, the training requirements for an accurate prior are more stringent. As such the practical feasibility of the filer design becomes an issue. This paper discusses the real time issues for large window filter design and how the bottlenecks were overcome to design practical large window multiresolution filters. The most crucial bottlenecks are the real memory required for training and the time required for training to obtain a satisfactory estimation of P(Y equals 1/x) or its prior for large windows. Among other improvements a method for data representation is developed that greatly reduces storage space for the large number of templates that occur for larger windows during the training of the filter. Parallel algorithms are designed that reduce hardware related time loss during training. In addition we take advantage of Bayesian filter methodology to train for large windows. While the algorithm works for larger windows, we demonstrate the feasibility of Bayesian multiresolution filter design for window sizes of up to 31 by 31.

The mean-absolute error of an increasing binary window-based filter can be expressed in terms of the errors of the individual erosions that comprise its erosion representation. This can be done recursively, so that the error of an m-erosion filter can be represented via errors of k-erosion filter can be represented via errors of k- erosion filters, where k < m. The error representation can be used for filter design form sample realizations by estimating the errors of individual erosions and then recursively computing the errors of multiple-erosion filters. Even for modestly large windows, the method is limited to obtaining filters composed of a relatively small number of erosions because the optimal (m + 1)-erosion filter need not be a superset of the optimal m-erosion filter. This paper considers two suboptimal search strategies to achieve faster design, and thereby permit filters comprised of larger numbers of erosions. A strict sequential search strategy proceeds under the assumption that the optimal (m + 1)-erosion filter is a superset of the optimal m-erosion filter. A look-ahead search strategy, in which erosion structuring elements are occasionally deleted from the growing basis, is close to a strict sequential search in computational efficiency, and can result in improved filtering over a full search.

Change detection generally is the difference between images. The differences or changes could be due to moving objects or a variation of illumination. In general the goal is to extract only changes due to moving object that occur int eh scene, and to ignore changes due to illumination. A requirement is that the detection has to be performed in real-time. The proposed change detection approach relies on a model assigning a vector to every pixel of the reference and the current image. Based on this model, linear independence is used to describe an operator for change detection. This previously published operator is based on the variance. The improved linear independence model consists in redefining the change detection operator to fulfill the real-time requirement and to improve the object detection performance. This model has been applied to surveillance and compared to the variance based linear independence detector. The operator proved to be robust to background and illumination changes and in the same time it detected object changes.

Some technical applications need a fast and reliable OCR for critical circumstances like low resolution and poor contrast. A concrete example is the real-time quality inspection system of Austrian banknotes. One requirement to the system is that it has to read two serial numbers on each banknote and to check if they are identical To solve the problem we have developed a novel method based an idea similar to pattern matching. However, instead of comparing entire images we use reduced sets of pixels, one for each different numeral. The detection is performed by matching these pixel sets with the corresponding pixels in the image being analyzed. We present an algorithm based on two cost functions that computes in a reasonable time the reduced pixel sets from a given set of image templates. The efficiency of our OCR has been increased considerably by introducing an appropriate set of image preprocessing operations. These are tailored especially to images with low resolution and poor contrast, bu they are simple enough to allow a fast real-time implementation. They can be seen as a normalization step that improves the image properties which are essential for pattern matching.

This paper presents the real-time vision system developed by Pavia University within the ENEA RAS project for the automatic driving of an intelligent snowcat able to follow the traces produced by other snowcats. A camera is used to acquire images of the scene in real-time; the image sequence is analyzed by a computer vision system which identifies the traces and produces a high level description of the scene. A further optional representation, in which black markers are superimposed onto the original acquired image, is transmitted to a human supervisor, located off-board.

Using teal-time image processing we have demonstrated a low bit-rate free-space optical communication system at a range of more than 20km with an average optical transmission power of less than 2mW. The transmitter is an autonomous one cubic inch microprocessor-controlled sensor node with a laser diode output. The receiver is a standard CCD camera with a 1-inch aperture lens, and both hardware and software implementations of the video semaphore decoding algorithm. With this system sensor data can be reliably transmitted 21 km form San Francisco to Berkeley.

This paper presents a new method for seamless tracking a moving object by using multiple fixed cameras controlled by a single processor. For seamless tracking a moving object with multiple cameras, we should extract important features of the object from input images captured by multiple cameras. Because two adjacent cameras, in general, have different parameters, such as: zooming ratio, color distribution, and direction of view, we use the color information of overlapped region and feature vector, such as object's ratio and moving direction. We first investigate the distribution of color component in the overlapped region. We then use the results to discriminate the target object and to hand over the current view. Experimental results show the feasibility of the proposed seamless tracking method and its real-time applications.

This paper presents a real-time head tracking method with a single black/white camera. This method mainly consists of two steps: block-based head tracking, and geometry-based correction. The block-based tracking method only uses low- level image information and can be used to track the free motion of the head. To improve the performance of block- based tracking during person turning around, we introduce a cylindrical model to approximate head, and extract features in the warped cylindrical surface. The correction step is introduced to address the shift caused by error accumulating of the block-based tracking method. This step utilizes the head contour geometry information, and measures the displacements between the head contour model and real contour in a current image, and hereby to correct the results of block tracking. We also set up a tracking system with a 350 megahertz microcomputer base on the proposed method, which operates at 30 frames per second with the tracking window size of 120 by 180 pixels and 80 tracking features.

Fingerprint verification is one of the most widespread techniques of personal identification. This paper describes the design of am minutiae-based fingerprint verification system including image preprocessing, feature extraction and fingerprint matching. Image preprocessing comprises the extraction of the region of interest, ridge segmentation, and ridge thinning. The features extracted from a fingerprint include fingerprint minutiae, i.e. ridge endings, and ridge bifurcations, as well as other related characteristics meant to improve the matching performances. The list of attributes of each minutia is extended with a feature vector, that resembles information extracted from the neighborhood region of the minutia. A measure of similarity between two minutiae can be expressed in terms of the distance between their corresponding feature vectors. We investigate two matching techniques based on the new approach of similar minutiae detection. A database containing 168 fingerprint images is used for experiments. The results reveal that the proposed system can achieve a good verification accuracy on this database. In addition, the proposed system meets the time requirements of practical acceptability as long as the average time for a verification is below 1.2 sec on a Sun ULTRA 1 workstation.

Transmitting huge geometry data over the network would bring severe delays that would effect the characteristics of reality and real-time of distributed virtual environment. To solve this problem we implemented a demand-driven progressive geometry transmission strategy with a prototype multi-user distributed virtual environment in client-server mode. Our data transmission strategy incorporates several techniques including multi-layer AOI management, demand- driven pre-fetching, progressive mesh based transmission and degraded presentation. The experimental result indicate that this strategy optimally utilizes the limited network bandwidth, efficiently saves memory space on local host, at the same time still meets with the real-time demand of virtual environment. In addition, to avoid the server being the system bottleneck, we designed and implemented a divisional data management mechanism with a demand-driven load-balancing algorithm.