Fast computation of 3D deformation fields is critical to bringing the application of automated elastic image registration algorithms to routine clinical practice. However, it lies beyond the computational power of current microprocessors; therefore requiring implementations using either massively parallel computers or application-specific hardware accelerators. The use of massively parallel computers in a clinical setting is not practical or cost-effective, therefore making the use of hardware accelerators necessary. We present a hardware pipeline that allows accelerating the computation of 3D deformation fields to speeds up to two orders of magnitude faster than software implementations on current workstations and about 64 times faster than other previously reported architectures. The pipeline implements a version of the free-form deformation calculation algorithm, which is optimized to minimize the number of arithmetic operations required to calculate the transformation of a given set of neighboring voxels, thereby achieving an efficient and compact implementation in hardware which allows its use as part of a larger system.

The use of correlation-based techniques to perform stereoscopic matching enables real-time frame rates to be achieved. The disparity maps produced are not as accurate as those produced using more sophisticated methods. This paper presents three ideas aimed at decreasing the computation time of standard correlation-based techniques. The reduction in computation time is achieved using two different methods. Firstly, the number of comparisons is reduced using MPEG-2 motion vectors to narrow the disparity search range to an optimised region and identifying areas of a scene which have remained static between frames so the previous disparity can be used. The second approach is an implementation of stereo using the above ideas on a GPU. The increases in error and frame rate can be controlled depending on the requirements of the application.

X-ray image visualized in real-time plays an important role in clinical applications. The real-time system design requires that images with the highest perceptual quality be acquired while minimizing the x-ray dose to the patient, which can result in severe noise that must be reduced. The approach based on the wavelet transform has been widely used for noise reduction. However, by removing noise, high frequency components belonging to edges that hold important structural information of an image are also removed, which leads to blurring the features. This paper presents a new method of x-ray image denoising based on fast lifting wavelet thresholding for general noise reduction and spatial filtering for further denoising by using a derivative model to preserve edges. General denoising is achieved by estimating the level of the contaminating noise and employing an adaptive thresholding scheme with variance analysis. The soft thresholding scheme is to remove the overall noise including that attached to edges. A new edge identification method of using approximation of spatial gradient at each pixel location is developed together with a spatial filter to smooth noise in the homogeneous areas but preserve important structures. Fine noise reduction is only applied to the non-edge parts, such that edges are preserved and enhanced. Experimental results demonstrate that the method performs well both visually and in terms of quantitative performance measures for clinical x-ray images contaminated by natural and artificial noise. The proposed algorithm with fast computation and low complexity provides a potential solution for real-time applications.

Denoising of medical images in wavelet domain has potential application in transmission technologies such as teleradiology. This technique becomes all the more attractive when we consider the progressive transmission in a teleradiology system. The transmitted images are corrupted mainly due to noisy channels. In this paper, we present a new real time image denoising scheme based on limited restoration of bit-planes of wavelet coefficients. The proposed scheme exploits the fundamental property of wavelet transform - its ability to analyze the image at different resolution levels and the edge information associated with each sub-band. The desired bit-rate control is achieved by applying the restoration on a limited number of bit-planes subject to the optimal smoothing. The proposed method adapts itself to the preference of the medical expert; a single parameter can be used to balance the preservation of (expert-dependent) relevant details against the degree of noise reduction. The proposed scheme relies on the fact that noise commonly manifests itself as a fine-grained structure in image and wavelet transform allows the restoration strategy to adapt itself according to directional features of edges. The proposed approach shows promising results when compared with unrestored case, in context of error reduction. It also has capability to adapt to situations where noise level in the image varies and with the changing requirements of medical-experts. The applicability of the proposed approach has implications in restoration of medical images in teleradiology systems. The proposed scheme is computationally efficient.

High-level Video content analysis such as video-surveillance is often limited by computational aspects of automatic image understanding, i.e. it requires huge computing resources for reasoning processes like categorization and huge amount of data to represent knowledge of objects, scenarios and other models. This article explains how to design and develop a "near real-time adaptive image datamart", used, as a decisional support system for vision algorithms, and then as a mass storage system. Using RDF specification as storing format of vision algorithms meta-data, we can optimise the data warehouse concepts for video analysis, add some processes able to adapt the current model and pre-process data to speed-up queries. In this way, when new data is sent from a sensor to the data warehouse for long term storage, using remote procedure call embedded in object-oriented interfaces to simplified queries, they are processed and in memory data-model is updated. After some processing, possible interpretations of this data can be returned back to the sensor. To demonstrate this new approach, we will present typical scenarios applied to this architecture such as people tracking and events detection in a multi-camera network. Finally we will show how this system becomes a high-semantic data container for external data-mining.

Human vigilance is limited; hence, automatic motion and distance detection is one of the central issues in video surveillance. Hereby, many aspects are of importance, this paper specially addresses: efficiency, achieving real-time performance, accuracy, and robustness against various noise factors. To obtain fully controlled test environments, an artificial development center for robot navigation is introduced in which several parameters can be set (e.g., number of objects, trajectories and type and amount of noise). In the videos, for each following frame, movement of stationary objects is detected and pixels of moving objects are located from which moving objects are identified in a robust way. An Exact Euclidean Distance Map (E²DM) is utilized to determine accurately the distances between moving and stationary objects. Together with the determined distances between moving objects and the detected movement of stationary objects, this provides the input for detecting unwanted situations in the scene. Further, each intelligent object (e.g., a robot), is provided with its E²DM, allowing the object to plan its course of action. Timing results are specified for each program block of the processing chain for 20 different setups. So, the current paper presents extensive, experimentally controlled research on real-time, accurate, and robust motion detection for video surveillance, using E²DMs, which makes it a unique approach.

Transit studies are important for planning a road network with optimal vehicular flow. A vehicular count is essential. This article presents a vehicle counting system based on video processing. An advantage of such system is the greater detail than is possible to obtain, like shape, size and speed of vehicles. The system uses a video camera placed above the street to image transit in real-time. The video camera must be placed at least 6 meters above the street level to achieve proper acquisition quality. Fast image processing algorithms and small image dimensions are used to allow real-time processing. Digital filters, mathematical morphology, segmentation and other techniques allow identifying and counting all vehicles in the image sequences. The system was implemented under Linux in a 1.8 GHz Pentium 4 computer. A successful count was obtained with frame rates of 15 frames per second for images of size 240x180 pixels and 24 frames per second for images of size 180x120 pixels, thus being able to count vehicles whose speeds do not exceed 150 km/h.

Auto white balancing (AWB) involves the process of making white colors to appear as white under different illuminants in digital imaging products such as digital still cameras. This paper presents a computationally efficient auto white balancing algorithm for real-time deployment in imaging products. The algorithm utilizes DWT (discrete wavelet transform) to perform multi-scale clustering (MSC), thus generating a computationally efficient implementation of the original MSC algorithm. The paper also discusses the steps taken to allow running this algorithm in real-time on a digital camera processor. The results of an actual implementation on the Texas Instruments TMS320DM320 processor are provided to illustrate the effectiveness of this algorithm in identifying an illuminant as compared to the widely used gray-world auto white balancing algorithm.

In this paper we present an enhanced real-time selective antialiasing solution. We propose to use a directional filtering technique as an antialiasing tool. The best post-processing antialiasing effect will be obtained if we apply the low-pass filter along local orientation of antialiased features. Previously authors proposed a complicated curve fitting method as a solution for the local feature antialiasing. Here we propose a more simple and efficient solution. Instead of using a curve fitting method based on second order intensity derivatives, we can use directly a set of first order derivatives applied on the z-buffer content. For each feature direction detected an appropriate directional Gaussian convolution filter can be applied. This way the lowpass filter is applied along local features selected for antialiasing, filtering out high frequency distortions due to intermodulation. In this approach the high-pass convolution filtering applied on the zbuffer has a twofold application: it selects the objects edges that need to be antialiased and it gives a local feature direction allowing for edge reconstruction. The advantage of the approach proposed here is that it preserves texture details. Textures usually are filtered independently using trilinear or anisotropic filtering, which with traditional antialiasing techniques leads to overblurring.

Eye detection plays a central role in an automatic face detection systems and it is also important for face recognition and face tracking. In this paper, we present a novel, unsupervised scheme for detecting eyes in skin patches based on our previous work. Working on the normalized RGB color space (NRGB ), we use a combination of corner identification, color and edges as heuristics for detecting eyes. Experimental results show that our scheme is very fast in the AR and Champion databases, while retaining very high detection rates.

This paper introduces an adaptive approach for image scaling. In addition, we present an efficient VLSI architecture to implement the proposed algorithm in hardware. The proposed architecture is designed to address the real-time constrain for high performance consumer products. A case study for printer application is presented.

Requirements for contemporary print inspection systems for industrial applications include, among others, high throughput, examination of fine details of the print, and inspection from various perspectives and different spectral sensitivity. Therefore, an optical inspection system for such tasks has to be equipped with several high-speed/high-resolution cameras, each acquiring hundreds of megabytes of data per second. This paper presents an inspection system which meets the given requirements by exploiting data parallelism and algorithmic parallelism. This is achieved by using complex field-programmable gate arrays (FPGA) for image processing. The scalable system consists of several processing modules, each representing a pair of a FPGA and a digital signal processor. The main chapters of the paper focus on the functionality implemented in the FPGA. The image processing algorithms include flat-field correction, lens distortion correction, image pyramid generation, neighborhood operations, a programmable arithmetic unit, and a geometry unit. Due to shortage of on-chip memory, a multi-port memory concept for buffering streams of data between off-chip and on-chip memories is used. Furthermore, performance measurements of the processing module are presented.

The demand on digital signal processing has been growing continuously driven by the increased data array size and sophisticated algorithms. The existing solutions using FPGA and/or ASIC have its advantages and drawbacks. The newly developed FPOA technology provides a flexible and effective programmable solution to meet the computation requirements and greatly reduces the schedule on development. This paper describes a flexible implementation of a complete digital signal processing on FPOA. An example space satellite application has been built and demonstrated on hardware.

To improve the weak target detection ability in radar applications a pulse compression is usually used that in the case linear FM modulation can improve the SNR. One drawback in here is that it can add the range side-lobes in reflectivity measurements. Using weighting window processing in time domain it is possible to decrease significantly the side-lobe level (SLL) and resolve small or low power targets those are masked by powerful ones. There are usually used classical windows such as Hamming, Hanning, etc. in window processing. Additionally to classical ones in this paper we also use a novel class of windows based on atomic functions (AF) theory. For comparison of simulation and experimental results we applied the standard parameters, such as coefficient of amplification, maximum level of side-lobe, width of main lobe, etc. To implement the compression-windowing model on hardware level it has been employed FPGA. This work aims at demonstrating a reasonably flexible implementation of FM-linear signal, pulse compression and windowing employing FPGA's. Classical and novel AF window technique has been investigated to reduce the SLL taking into account the noise influence and increasing the detection ability of the small or weak targets in the imaging radar. Paper presents the experimental hardware results of windowing in pulse compression radar resolving several targets for rectangular, Hamming, Kaiser-Bessel, (see manuscript for formula) functions windows. The windows created by use the atomic functions offer sufficiently better decreasing of the SLL in case of noise presence and when we move away of the main lobe in comparison with classical windows.

We describe in this article a new multi-view auto-stereoscopic display system with a real time architecture to generate images of n different points of view of a 3D scene. This architecture generates all the different points of view with only one generation process, the different pictures are not generated independently but all at the same time. The architecture generates a frame buffer that contains all the voxels with their three dimensions and regenerates the different pictures on demand from this frame buffer. The need of memory is decreased because there is no redundant information in the buffer.

In this article we present the local operations in image processing based upon spatial 2D discrete convolution. We study different implementation of such local operations. We also present the principles and the design flow of the AAA methodology and its associated CAD software tool for integrated circuit (SynDEx-IC). In this methodology, the algorithm is modeled by Conditioned (if - then - else) and Factorized (Loop) Data Dependence Graph and the optimized implementation is obtained by graph transformations. The AAA/SynDEx-IC is used to specify and to optimize the some digital image filters on FPGA XC2100 board.

An efficient real-time video stream enhancement algorithm based on illuminance-reflectance model is proposed for improving the visual quality of digital video streams captured under insufficient and/or non-uniform lighting conditions. The paper presents computational methods for estimation of scene illuminance and reflectance, adaptive dynamic range compression of illuminance, and adaptive enhancement for mid-tone frequency components. This algorithm is an effective and efficient technique for image enhancement with relatively simple computational procedures, which makes real-time enhancement of digital videos successfully realized. It also demonstrates strong robustness and high image quality when compared to other techniques.

This paper presents an improved parallel Two-Pass Hexagonal (TPA) algorithm constituted by Linear Hashtable Motion Estimation Algorithm (LHMEA) and Hexagonal Search (HEXBS) for motion estimation. Motion Vectors (MV) are generated from the first-pass LHMEA and used as predictors for second-pass HEXBS motion estimation, which only searches a small number of Macroblocks (MBs). We used hashtable into video processing and completed parallel implementation. The hashtable structure of LHMEA is improved compared to the original TPA and LHMEA. We propose and evaluate parallel implementations of the LHMEA of TPA on clusters of workstations for real time video compression. The implementation contains spatial and temporal approaches. The performance of the algorithm is evaluated by using standard video sequences and the results are compared to current algorithms.

In this paper we present the capability and real-time processing features of a new type of L-filter for the removal of impulsive and multiplicative noise in real-time image processing applications. The proposed filter uses the robust RMestimator in the filtering scheme of L-filter according with the RM-KNN filtering algorithms. Extensive simulation results in known reference images, and SAR images have demonstrated that the proposed filter consistently outperforms other filters by balancing the tradeoff between noise suppression and detail preservation. The criteria used to compare the performance or various filters were the PSNR, MAE, and processing time. The real-time implementation of proposed algorithm was realized on the DSP TMS320C6701. The processing time of proposed filter includes the time of data acquisition, processing and store data. We found that the processing time values of proposed filter depends of the image to process and do not practically vary for different complex noise level; these values depend also of the calculation of influence functions, parameters of the proposed filter, and different distribution functions used to calculate the coefficients of the new type of L-filter.

Recently, monitoring cameras for security have been extensively increasing. However, it is normally difficult to know when and where we are monitored by these cameras and how the recorded images are stored and/or used. Therefore, how to protect privacy in the recorded images is a crucial issue. In this paper, we address this problem and introduce a framework for security monitoring systems considering the privacy protection. We state requirements for monitoring systems in this framework. We propose a possible implementation that satisfies the requirements. To protect privacy of recorded objects, they are made invisible by appropriate image processing techniques. Moreover, the original objects are encrypted and watermarked into the image with the "invisible" objects, which is coded by the JPEG standard. Therefore, the image decoded by a normal JPEG viewer includes the objects that are unrecognized or invisible. We also introduce in this paper a so-called "special viewer" in order to decrypt and display the original objects. This special viewer can be used by limited users when necessary for crime investigation, etc. The special viewer allows us to choose objects to be decoded and displayed. Moreover, in this proposed system, real-time processing can be performed, since no future frame is needed to generate a bitstream.

This paper presents algorithms for image processing and image analysis of wood piece materials. The algorithms were designed for auto-detection of wood piece materials on a moving conveyor belt or a truck. When wood objects on moving, the hard task is to trace the contours of the objects in n optimal way. To make the algorithms work efficiently in the plant, a flexible online system was designed and developed, which mainly consists of image acquisition, image processing, object delineation and analysis. A number of newly-developed algorithms can delineate wood objects with high accuracy and high speed, and in the wood piece analysis part, each wood piece can be characterized by a number of visual parameters which can also be used for constructing experimental models directly in the system.

Image similarity-based image registration is an iterative process that, depending on the number of degrees of freedom in the underlying transformation, may require hundreds to tens of thousands of image similarity computations to converge on a solution. Computation time often limits the use of such algorithms in real-life applications. We have previously shown that hardware acceleration can significantly reduce the time required to register two images. However, the hardware architectures we presented were limited to mutual information calculation, which is one of several commonly used image similarity measures. In this article we show how our architecture can be adapted for the calculation of other image similarity measures in approximately the same time and using the same hardware resources as those for the mutual information case. As in the case of mutual information calculation, the joint histogram is calculated as a first step. The image similarity measures considered are mutual information, normalized mutual information, normalized cross correlation, mean-square sum of differences and ratio image uniformity. We show how all these image similarities can be calculated from the joint histogram in a small fraction of the time required to calculate the joint histogram itself.

The goal of this project was to detect the intensity of traffic on a road at different times of the day during daytime. Although the work presented utilized images from a section of a highway, the results of this project are intended for making decisions on the type of intervention necessary on any given road at different times for traffic control, such as installation of traffic signals, duration of red, green and yellow lights at intersections, and assignment of traffic control officers near school zones or other relevant locations. In this project, directional patterns are used to detect and count the number of cars in traffic images over a fixed area of the road to determine local traffic intensity. Directional patterns are chosen because they are simple and common to almost all moving vehicles. Perspective vision effects specific to each camera orientation has to be considered, as they affect the size and direction of patterns to be recognized. In this work, a simple and fast algorithm has been developed based on horizontal directional pattern matching and perspective vision adjustment. The results of the algorithm under various conditions are presented and compared in this paper. Using the developed algorithm, the traffic intensity can accurately be determined on clear days with average sized cars. The accuracy is reduced on rainy days when the camera lens contains raindrops, when there are very long vehicles, such as trucks or tankers, in the view, and when there is very low light around dusk or dawn.