Showing papers in "Real-time Imaging in 2003"
TL;DR: This paper provides a detailed survey of the various studies in areas related to the tracking of people and body parts such as face, hands, fingers, legs, etc., and modeling behavior using motion analysis.
Abstract: Video analysis of human dynamics is an important area of research devoted to detecting people and understanding their dynamic physical behavior in a complex environment that can be used for biometric applications. This paper provides a detailed survey of the various studies in areas related to the tracking of people and body parts such as face, hands, fingers, legs, etc., and modeling behavior using motion analysis.
182 citations
TL;DR: The developed rule-based approach to passive auto-focusing for digital still cameras is compared to the commonly used global search and binary search algorithms in terms of focusing speed and power consumption and it is shown that the introduced rule- based search algorithm achieves a lower number of focusing iterations or faster focusing speed, and a higher number of steps or lower power consumption.
Abstract: This paper presents the development and real-time implementation of a rule-based approach to passive auto-focusing for digital still cameras. The implementation is performed on the processor DM310 which is specifically manufactured by Texas Instruments for digital still cameras. The squared-gradient sharpness function is considered to measure the amount of high-frequency content in an out-of-focus image. This sharpness function is then used within a rule-based search algorithm to obtain the focused lens position via moving the lens head step-motor. The developed rule-based approach is compared to the commonly used global search and binary search algorithms in terms of focusing speed and power consumption. It is shown that the introduced rule-based search algorithm achieves a lower number of focusing iterations or faster focusing speed, and a lower number of steps or lower power consumption.
132 citations
TL;DR: A real-time face detection algorithm for locating faces in images and videos that finds not only the face regions, but also the precise locations of the facial components such as eyes and lips using a simple quadratic polynomial model.
Abstract: This paper presents a real-time face detection algorithm for locating faces in images and videos. This algorithm finds not only the face regions, but also the precise locations of the facial components such as eyes and lips. The algorithm starts from the extraction of skin pixels based upon rules derived from a simple quadratic polynomial model. Interestingly, with a minor modification, this polynomial model is also applicable to the extraction of lips. The benefits of applying these two similar polynomial models are twofold. First, much computation time are saved. Second, both extraction processes can be performed simultaneously in one scan of the image or video frame. The eye components are then extracted after the extraction of skin pixels and lips. Afterwards, the algorithm removes the falsely extracted components by verifying with rules derived from the spatial and geometrical relationships of facial components. Finally, the precise face regions are determined accordingly. According to the experimental results, the proposed algorithm exhibits satisfactory performance in terms of both accuracy and speed for detecting faces with wide variations in size, scale, orientation, color, and expressions.
122 citations
TL;DR: A new class of filters for noise attenuation is introduced and its relationship with commonly used filtering techniques is investigated and it is indicated that the new filter outperforms the VMF, as well as other techniques currently used to eliminate impulsive noise in color images.
Abstract: In this paper, we address the problem of impulsive noise reduction in multichannel images. A new class of filters for noise attenuation is introduced and its relationship with commonly used filtering techniques is investigated. The computational complexity of the new filter is lower than that of the vector median filter (VMF). Extensive simulation experiments indicate that the new filter outperforms the VMF, as well as other techniques currently used to eliminate impulsive noise in color images.
112 citations
TL;DR: An extensive set of methods for segmentation of uncompressed video sequences are reviewed and classify them in several parts, depending on the information used to detect shot changes.
Abstract: We present in this paper a review of methods for segmentation of uncompressed video sequences. Video segmentation is usually performed in the temporal domain by shot change detection. In case of real-time segmentation, computational complexity is one of the criteria which has to be taken into account when comparing different methods. When dealing with uncompressed video sequences, this criterion is even more significant. However, previous published reviews did not involve complexity criterion when comparing shot change detection methods. Only recognition rate and ability to classify detected shot changes were considered. So contrary to previous reviews, we give here the complexity of most of the described methods. We review in this paper an extensive set of methods presented in the literature and classify them in several parts, depending on the information used to detect shot changes. The earliest methods were comparing successive frames by relying on the most simple elements, that is to say pixels. Comparison could be performed on a global level, so methods based on histograms were also proposed. Block-based methods have been considered to process data at an intermediate level, between local (using pixels) and global (using histograms) levels. More complex features can be involved, resulting in feature-based methods. Alternatively some methods rely on motion as a criterion to detect shot changes. Finally, different kinds of information could be combined together in order to increase the quality of shot change detection. So our review will detail segmentation methods based on the following information: pixel, histogram, block, feature, motion, or other kind of information.
96 citations
TL;DR: A new image watermarking scheme which is robust to RST attacks with cropping is proposed by improving Fourier-Mellin transform based watermarked (FMW) for improvement of realization and performance.
Abstract: This paper proposes a new image watermarking scheme which is robust to RST attacks with cropping by improving Fourier-Mellin transform based watermarking (FMW). The proposed scheme reorders and modifies function blocks of FMW for improvement of realization and performance. Unlike FMW, our method uses log-polar map (LPM) in the spatial domain for scaling invariance, while translation invariance is provided by the use of an invariant centroid (IC) as the origin of LPM. IC is a gravity center of a central area on gray scale image that is invariant although an image is attacked by RST. For this, its calculation method is proposed. Also since LPM includes the property which transforms rotation of Cartesian coordinates system into a cyclic shift, 2-D DFT is performed on the LPM image and the magnitude spectrum extracted to provide a domain that is rotation invariant. The resulting domain, which is invariant to RST, is then used as the watermark-embedding domain. Furthermore, to prevent the watermarked image from degrading due to the coordinate system conversion, only LPM image of watermark signal is inverse mapped to Cartesian coordinates and add to the original image. Experimental results demonstrate that the proposed scheme is robust to RST attacks.
73 citations
TL;DR: A method for paper label detection on polymer parts is introduced, aimed at enhancing the classification results by merging connected parts of an object.
Abstract: Near-infrared (NIR) spectroscopy is widely used in laboratory and industrial applications for material classification. While standard spectrometers only allow measurement at one sampling point at a time, NIR Spectral Imaging techniques can identify, in real-time, both the size and shape of an object as well as the material it is made from. The robust classification of materials, such as polymers, is based on their characteristic reflectance spectra. As a sample application, we present the real-time classification of waste polymers in a prototype of an automated industrial sorting facility. Sorting requires the correct material, size and shape of the entire object to be known for reliable separation. In this paper, a method for paper label detection on polymer parts is introduced, aimed at enhancing the classification results by merging connected parts of an object.
72 citations
TL;DR: This technique is used to estimate the most important compounds which play a role in the ripening of tomatoes, using the total compound concentrations and the spatial distribution of the concentrations as criteria.
Abstract: Independent Component Analysis is one of the most widely used methods for blind source separation. In this paper we use this technique to estimate the most important compounds which play a role in the ripening of tomatoes. Spectral images of tomatoes were analyzed. Two main independent components were found. These components resemble the actual absorption spectra of lycopene and chlorophyll. Concentration images of these compounds show increase of one compound and decrease of the other during ripening. The method can be implemented in an unsupervised real time sorting machine, using the total compound concentrations and the spatial distribution of the concentrations as criteria.
63 citations
TL;DR: The efficiency of the real-time surface inspection method is empirically evaluated in two different problems including textures from the Outex database and from a paper inspection problem.
Abstract: In this paper a real-time surface inspection method based on texture features is introduced. The proposed approach is based on the Local Binary Pattern (LBP) texture operator and the Self-Organizing Map (SOM). A very fast software implementation of the LBP operator is presented. The SOM is used as a powerful classifier and visualizer. The efficiency of the method is empirically evaluated in two different problems including textures from the Outex database and from a paper inspection problem.
58 citations
TL;DR: It is shown, that classifiers built in dissimilarity spaces may also be applied significantly faster than the nearest-neighbor rule, which is of crucial importance for real-time spectral imaging applications.
Abstract: For the sake of classification, spectra are traditionally represented by points in a high-dimensional feature space, spanned by spectral bands. An alternative approach is to represent spectra by dissimilarities to other spectra. This relational representation enables one to treat spectra as connected entities and to emphasize characteristics such as shape, which are difficult to handle in the traditional approach. Several classification methods for relational representations were developed and found to outperform the nearest-neighbor rule. Existing studies focus only on the performance measured by the classification error. However, for real-time spectral imaging applications, classification speed is of crucial importance. Therefore, in this paper, we focus on the computational aspects of the on-line classification of spectra. We show, that classifiers built in dissimilarity spaces may also be applied significantly faster than the nearest-neighbor rule.
56 citations
TL;DR: The experimental results indicate that the segmentation method preceded by spatial color nonuniformity correction accurately extracts color clusters with complex shapes and therefore correctly segments the inspected images, which can thus be used in industrial environments where real-time inspection of color objects is required.
Abstract: In pharmaceutical blister packing, it is today part of the recommended security standard to inspect tablets in each blister before it is sealed: In this paper an automated visual inspection system is described, which detects missing and broken tablets, tablet fragments, as well as the color, size, and shape of individual tablets. The system operates either in "training" or "inspection" mode. In training mode, the image of defect-free blisters is used to extract the blister model, which is composed of the spatial color nonuniformity correction function, positions of blisters, positions of tablets in blisters, the color labeling function, and position, size, and shape of each tablet and corresponding pre-specified tolerances. The blister model allows effective and real-time analysis of blisters in inspection mode. The most important parts of the system are correction of the adverse effects of spatial color nonuniformity and color segmentation. A method recently proposed for spatial intensity nonuniformity correction has been extended to suppress spatial color nonuniformity in color images. A novel nonparametric clustering-based segmentation method is proposed, which finds the valleys between color clusters. The experimental results indicate that the segmentation method preceded by spatial color nonuniformity correction accurately extracts color clusters with complex shapes and therefore correctly segments the inspected images. The automated visual inspection system can thus be used in industrial environments where real-time inspection of color objects is required.
TL;DR: It is demonstrated that foveation in the DCT domain can actually result in computational speed-ups, and can be incorporated into standard motion compensation and discrete cosine transform (DCT)-based video coding techniques for low bit rate video coding, without incurring prohibitive complexity overhead.
Abstract: Lossy video compression methods often rely on modeling the abilities and limitations of the intended receiver, the human visual system (HVS), to achieve the highest possible compression with as little effect on perceived quality as possible. Foveation, which is non-uniform resolution perception of the visual stimulus by the HVS due to the non-uniform density of photoreceptor cells in the eye, has been demonstrated to be useful for reducing bit rates beyond the abilities of uniform resolution video coders. In this work, we present real-time foveation techniques for low bit rate video coding. First, we develop an approximate model for foveation. Then, we demonstrate that foveation, as described by this model, can be incorporated into standard motion compensation and discrete cosine transform (DCT)-based video coding techniques for low bit rate video coding, such as the H.263 or MPEG-4 video coding standards, without incurring prohibitive complexity overhead. We demonstrate that foveation in the DCT domain can actually result in computational speed-ups. The techniques presented can be implemented using the baseline modes in the video coding standards and do not require any modification to, or post-processing at, the decoder.
TL;DR: It is shown that recursive filtering, which was introduced to approximate Gaussian convolution, can be extended to the space-variant case and leads to a very simple implementation that makes it ideal for that application.
Abstract: Animal visual systems have solved the problem of limited resources by allocating more processing power to central than peripheral vision. Foveation considerably reduces the amount of data per image by progressively decreasing the resolution at the periphery while retaining a sharp center of interest. This strategy has important applications in the design of autonomous systems for navigation, tracking and surveillance. Central to foveation is a space-variant Gaussian filtering scheme that gradually blurs out details as the distance to the image center increases. Unfortunately Gaussian convolution is a computationally expensive operation, which can severely limit the real-time applicability of foveation. In the space-variant case, the problem is even more difficult as traditional techniques such as the fast Fourier transform cannot be employed because the convolution kernel is different at each pixel. We show that recursive filtering, which was introduced to approximate Gaussian convolution, can be extended to the space-variant case and leads to a very simple implementation that makes it ideal for that application. Three main recursive algorithms have emerged, produced by independent derivation methods. We assess and compare their performance in traditional filtering applications and in our specific space-variant case. All three methods drastically cut down the cost of Gaussian filtering to a limited number of operations per pixel that is independent of the scale selected. In addition we show that two of those algorithms have excellent accuracy in that the output they produce differs from the output obtained performing real Gaussian convolution by less than 1%.
TL;DR: A set of recipes for implementation of the Kalman filter to a variety of real- time imaging settings, the presentation of a set of object-oriented requirements, and a design for a class of Kalman filters suitable for real-time image processing are presented.
Abstract: Kalman filters are an important technique for building fault-tolerance into a wide range of systems, including real-time imaging. From a software engineering perspective, however, it is not easy to build Kalman filters, Each has to be custom designed and most software engineers are not sufficiently grounded in the necessary systems theory to perform this design.The contributions of this paper, therefore, are a set of recipes for implementation of the Kalman filter to a variety of real-time imaging settings, the presentation of a set of object-oriented requirements, and a design for a class of Kalman filters suitable for real-time image processing.First, we describe the Kalman filter and motivate its use as a mechanism for fault-tolerant computing and sensor fusion. Next, the details of using Kalman filters in imaging applications are discussed and several associated algorithms presented. Then, the advantages of using object-oriented specification, design and languages for the implementation of Kalman filters are explored. Finally, we present a specification and design for a class of Kalman filters, which is suitable for coding. This work extends significantly upon that first appearing in 2003 at an SPIE conference (Laplante and Neill, proceedings of the real-time imaging conference, SPIE, Santa Clara, January 2003, pp. 22-29).
TL;DR: This paper reports the first real-time implementation of the complete visual attention mechanism on a compact and low-power architecture called ProtoEye, designed for general purpose low-level image processing.
Abstract: Visual attention is the ability to rapidly detect the visually salient parts of a given scene on which higher level vision tasks, such as object recognition, can focus. Found in biological vision, this mechanism represents a fundamental tool for computer vision. This paper reports the first real-time implementation of the complete visual attention mechanism on a compact and low-power architecture. Specifically, the saliency-based model of visual attention was implemented on a highly parallel single instruction, multiple data architecture called ProtoEye. Conceived for general purpose low-level image processing, ProtoEye consists of a 2D array of mixed analog-digital processing elements. To reach real-time, the operations required for visual attention computation were optimally distributed on the analog and digital parts. The currently available prototype runs at a frequency of 14 images/s and operates on 64 × 64 gray level images. Extensive testing and run-time analysis of the system stress the strengths of the architecture.
TL;DR: The problem of detecting a moving element in a sequence of images is transformed into the recognition of a pattern on a static image, namely the LSR footprint, which is proposed as the basic characteristic for the detection and subsequent classification of moving elements in image sequences.
Abstract: In this article, the length-speed ratio (LSR) is proposed as a basic characteristic for the real-time detection of moving objects. We define the LSR of a uniform moving zone as the relation between its length in the direction of motion and the speed of this motion. For a given zone of the image with uniform gray level (or patch), the greater its length in the direction of motion and the smaller its speed, the greater its LSR. A moving element is generally composed of various zones of uniform gray levels (or patches), which move with the same speed but which have different lengths in the direction of motion and which therefore have a characteristic set of LSR values. In this article, this "LSR footprint" is proposed as the basic characteristic for the detection and subsequent classification of moving elements in image sequences. The problem of detecting a moving element in a sequence of images is transformed into the recognition of a pattern on a static image, namely the LSR footprint. We also specify how to obtain this characteristic in real time, we discuss its invariants and we consider the cases for which LSR detection of movement is applicable. We also present its use in some significant examples and we compare it with other methods applicable to similar computational problems.
TL;DR: The experimental results show that the proposed edge detector can increase the resolution of an existing intelligent line camera by a factor of 6 or more and is effective for real-time, low-cost subpixel edge detection.
Abstract: In this paper we present a field programmable gate array (FPGA) implementation of a real-time subpixel edge detector. In comparison to existing edge detectors, the proposed method is implemented in hardware and its computational cost and complexity are very low. This in turn reduces the overall system cost quite significantly. The edge detection method is based on the approximation of the edge profile with a first-order linear function and then extrapolating the line to intercept a constant function. The position where both lines intersect will be the edge position. The detector is capable of processing a high-resolution CCD linear sensor at 2000 frames/s at the maximum clock rate of the sensor of 2 MHz. For that reason it is as fast as an analogue approach. The experimental results show that the proposed method can increase the resolution of an existing intelligent line camera by a factor of 6 or more. From the above it can be concluded that the proposed edge detector is effective for real-time, low-cost subpixel edge detection.
TL;DR: This work presents an efficient image-based rendering algorithm that generates views of a scene's photo hull that takes advantage of epipolar geometry to efficiently reconstruct the geometry and visibility of ascene.
Abstract: We present an efficient image-based rendering algorithm that generates views of a scene's photo hull. The photo hull is the largest 3D shape that is photo-consistent with photographs taken of the scene from multiple viewpoints. Our algorithm, image-based photo hulls (IBPH), like the image-based visual hulls (IBVH) algorithm from Matusik et al. on which it is based, takes advantage of epipolar geometry to efficiently reconstruct the geometry and visibility of a scene. Our IBPH algorithm differs from IBVH in that it utilizes the color information of the images to identify scene geometry. These additional color constraints result in more accurately reconstructed geometry, which often projects to better synthesized virtual views of the scene. We demonstrate our algorithm running in a realtime 3D telepresence application using video data acquired from multiple viewpoints.
TL;DR: A real-time implementation of an approximation of the support vector machine (SVM) decision rule is proposed, based on an improvement of a supervised classification method using hyperrectangles, which is useful for real- time image segmentation.
Abstract: A real-time implementation of an approximation of the support vector machine (SVM) decision rule is proposed. This method is based on an improvement of a supervised classification method using hyperrectangles, which is useful for real-time image segmentation. The final decision combines the accuracy of the SVM learning algorithm and the speed of a hyperrectangles-based method. We review the principles of the classification methods and we evaluate the hardware implementation cost of each method. We present the combination algorithm, which consists of rejecting ambiguities in the learning set using SVM decision, before using the learning step of the hyperrectangles-based method. We present results obtained using Gaussian distribution and give an example of image segmentation from an industrial inspection problem. The results are evaluated regarding hardware cost as well as classification performances.
TL;DR: A fast and bounded method to optimize the required network resources (bandwidth) to guarantee a maximum deterministic delay for a given workload and network state and is shown to be practically as efficient (or even better) than the best known RC-EDF policies with several nodes.
Abstract: The transmission of video requires new service models for providing quality of service (QoS). Some of these models are based on resource reservation and admission control while others rely on resource provisioning. In both approaches, optimally estimating the resource requirements of a given video is a key issue, since they are always very demanding. This paper introduces a fast and bounded method to optimize the required network resources (bandwidth) to guarantee a maximum deterministic delay for a given workload and network state. The method is based on characterizing the workload of a stored video using a reduced set of points obtained from an off-line analysis of its empirical envelope. The paper also proposes the above workload characterization to perform the admission control test using a WFQ scheduler. The evaluation of this scheme shows that it can achieve an utilization higher than a 50% for a single flow with a 1-s deadline and reach up to an 80% with a 5-s delay. Results are also compared to tests based on EDF schemes, which have been proved to be optimal with one node. It is shown that the proposed scheme is a little less efficient than the optimal EDF scheduler with one node (as expected), but it is practically as efficient (or even better with a moderately high number of nodes) than the best known RC-EDF policies with several nodes. This is an interesting result because it shows that the WFQ schedulers can achieve a similar efficiency to EDF schedulers avoiding the complexity of their admission control tests.
TL;DR: Real-time considerations related to the multi-processor architecture and the multitasking operating system that allow the implementation of the proposed architecture for real-time vision systems applications are emphasized.
Abstract: PC-based real-time vision systems are becoming a de facto standard in industrial applications. They are composed of an illumination system, an image acquisition system and a processing system. In this article, a modular and scalable architecture for real-time vision systems is proposed. On the one hand, we define an acquisition module that allows simultaneous acquisition of up to three monochrome cameras. The acquisition system can be scaled by adding more acquisition modules. The architecture allows simultaneous acquisition of all the modules. On the other hand, we define a processing system composed of different modules and sub-modules which specialize in a particular task: the master module interacts with the external systems; the slave module interacts with the acquisition system and manages the results of its processing sub-modules; each processing sub-module processes the information provided by one single camera. Scalability is provided by increasing the number of slave modules and processing submodules that compose the complete vision system. Fast real-time applications can be achieved by assigning one processor per processing sub-module. In this case, multiple PC can be used. Inter-computer communication among modules is carried out by means of sockets (following a maste-slave design pattern); intra-computer communication is carried out by means of pipes, shared memory and events. We emphasize herein some real-time considerations related to the multi-processor architecture and the multitasking operating system that allow the implementation of the proposed architecture for real-time vision systems applications. An implementation of this architecture is exemplified with an application successfully installed in a manufacturing company.
TL;DR: Dynamic Reconfiguration (or Run-Time Reconfigured), a technique based on the reuse of the same device (an FPGA configured on the fly) by scheduling the execution of different algorithms building an application is discussed.
Abstract: During the last few years, many architectures using processors and/or field programmable gate arrays (FPGA) were built to accelerate computationally complex problems. The processors allow a high degree of flexibility, whilst the FPGA implementation might be considerably faster. In spite of the possibility of reconfiguring the conventional FPGA an unlimited number of time, many of these architectures were built to compute a single application. If the FPGA is reconfigured several times to execute various algorithms, the configuration time increases and degrades global performances. In this paper, an architecture dedicated to real-time image processing using the AT40K reconfigurable FPGA family is presented (ARDOISE project 1 ). We discuss Dynamic Reconfiguration (or Run-Time Reconfiguration), a technique based on the reuse of the same device (an FPGA configured on the fly) by scheduling the execution of different algorithms building an application. The techniques and the tools developed to test and use the system are described.
TL;DR: Experimental results show that the proposed real-time restoration algorithm can enhance the input video much better than simple filtering techniques and is relatively insensitive to the human visual system.
Abstract: A novel framework of real-time video enhancement is proposed. The proposed framework is based on the regularized iterative image restoration algorithm, which iteratively removes degradation effects under a priori constraints. Although regularized iterative image restoration is proven to be a successful technique in restoring degraded images, its application is limited within still images or off-line video enhancement because of its iterative structure. In order to enable this iterative restoration algorithm to enhance the quality of video in real-time, each frame of video is considered as the constant input and the processed previous frame is considered as the previous iterative solution. This modification is valid only when the input of the iteration, that is each frame, remains unchanged throughout the iteration procedure. Because every frame of general video sequence is different from each other, each frame is segmented into two regions: still background and moving objects. These two regions are processed differently by using a segmentation-based spatially adaptive restoration and a background generation algorithms. Experimental results show that the proposed real-time restoration algorithm can enhance the input video much better than simple filtering techniques. The proposed framework enables real-time video enhancement at the cost of image quality only in the moving object area of dynamic shots, which is relatively insensitive to the human visual system.
TL;DR: A new method is presented for automatic correction of distortions and for spectral calibration (which band corresponds to which wavelength) of spectral images recorded by means of a spectrograph using a penalized likelihood method in a quasi-Newton iterative optimization technique.
Abstract: A new method is presented for automatic correction of distortions and for spectral calibration (which band corresponds to which wavelength) of spectral images recorded by means of a spectrograph. The method consists of recording a bar-like pattern with an illumination source with spectral bands (e.g. neon or mercury). Using prior information of the wavelength of these spectral bands and the spatial arrangement of the bars, a template image is constructed where the spectral axis is linearly related with wavelength. Next, a grid is posed on both the recorded and template image. Using a penalized likelihood method in a quasi-Newton iterative optimization technique, points of the grid on the recorded image are shifted such that the transformed (warped) image has a high resemblance (likelihood) to the template image and a low distortion (penalty term). The method is fully automatic and does not require any landmark extraction. After the transformation grid has been established, every new recorded image can be corrected in real time for any spectral and spatial distortion using fast bilinear interpolation. Recalibration of the system can be done reasonably fast using a previously calculated grid.
TL;DR: The paper addresses optimization techniques available by the application of the SPOT, a new development tool based on visualization of the scheduled assembly code by a two-dimensional interactive schedule editor equipped with feedback mechanisms deduced from analysis of data dependencies and resource allocation conflicts.
Abstract: Although the hardware platform is often seen as the most important element of real-time imaging systems, software optimization can also provide remarkable reduction of overall computational costs. The recommended code development flow for digital signal processors based on the TMS320C6000TM architecture usually involves three phases: development of C code, refinement of C code, and programming linear assembly code. Each step requires a different level of knowledge of processor internals. The developer is not directly involved in the automatic scheduling process. In some cases, however, this may result in unacceptable code performance. A better solution can be achieved by scheduling the assembly code by hand. Unfortunately, scheduling of software pipelines by hand not only requires expert skills but is also time consuming, and moreover, prone to errors. To overcome these drawbacks we have designed an innovative development tool--the Software Pipeline Optimization Tool (SPOTTM). The SPOT is based on visualization of the scheduled assembly code by a two-dimensional interactive schedule editor, which is equipped with feedback mechanisms deduced from analysis of data dependencies and resource allocation conflicts. The paper addresses optimization techniques available by the application of the SPOT. Furthermore, the benefit of the SPOT is documented by more than 20 optimized image processing algorithms.
TL;DR: An efficient color edge-based change detection scheme (CECD) is utilized in this paper to meet the real-time performance for HSS and can achieve better spatial accuracy and temporal coherency than COST211 AM.
Abstract: MPEG-4 introduces the concept of video object to support content-based functionalities. Video object segmentation is a key step in defining the contents of any video sequences. Head-shoulder sequence (HSS) is typical in video conferencing and surveillance systems, in which real-time performance is required. Since background information can be obtained in advance and pre-stored, video segmentation for HSS can use background information a priori. To avoid the critical selection of threshold for gradient-based method, and to overcome the insufficiency of monochrome intensity-based change detection, an efficient color edge-based change detection scheme (CECD) is utilized in this paper. In order to meet the real-time performance for HSS, it is implemented in the cellular neural networks (CNN) architecture. The algorithm is mainly based on 3 by 3, linear templates. Because of CNN's high parallelism and computational abilities, real-time performance is achieved. Experimental results on several test sequences show the robustness of this approach. It can achieve better spatial accuracy and temporal coherency than COST211 AM.
TL;DR: The 2D algorithm is found to improve the performance of digital filtering systems by segmenting the 2D input into smaller block sizes, which are shown to be efficient and lead to highly parallel implementation.
Abstract: A simple and efficient design and implementation of the two-dimensional (2D) parallel block-filtering algorithm by fast number theoretic transforms is presented. The algorithm is generalized to accommodate any filter size for optimum segmentation of the input data in order to achieve efficient filtering and optimal parallelization. The 2D algorithm is found to improve the performance of digital filtering systems by segmenting the 2D input into smaller block sizes, which are shown to be efficient and lead to highly parallel implementation. In this paper the parallel architecture of the 2D block-filtering method is presented and the implementation of fast 2D block filtering using the 2D new Mersenne number transform (2D NMNT) for digital filtering is demonstrated on a multiprocessor platform. The mathematical derivations of the input stage, output stage and the direct 2D FIR filtering equation are also given. The algorithm's efficiency is tested, and new results are given showing an improved performance, as evidenced by the highly parallel algorithm structure and the use of smaller transform sizes.
TL;DR: This paper combines the simplicity of the rate-distortion (R-D) predictive models with the improved R-D performance offered by the iterative approaches and design effective rate control algorithms for low encoding delay and low complexity video compression.
Abstract: Rate control in standards such as MPEG-2 or H263 plays an important role in controlling the compression rate and stabilizing the decoding and playback quality. In this paper, we combine the simplicity of the rate-distortion (R-D) predictive models with the improved R-D performance offered by the iterative approaches and we design effective rate control algorithms for low encoding delay and low complexity video compression. The proposed algorithms initially utilize the Lagrangian formulation in order to add a local R-D optimization component into the current rate control scheme of the standard. Subsequently, two online controllers are designed based on this local optimization scheme for adjusting the bit rate produced by the standard. Extensive experiments show improvements over the MPEG-2 scheme of 0.5-1 dB per frame for a variety of test sequences when the same target bit rates are maintained.
TL;DR: This work examines best practices in software engineering, in the form of patterns and design principles, with reference to imaging systems and within the context of the Java imaging APIs.
Abstract: Imaging systems are traditionally developed using structured analysis and design techniques at best. Such approaches tend to be rigid with respect to changing needs, technologies, devices and algorithms--for example, when additional compression algorithms are needed or attached devices are changed large parts of software applications employing those techniques and interfacing with those devices must be modified to accommodate the change. In a larger perspective, these systems are difficult or impossible to reuse; each new problem requires a new solution.This is generally undesirable and often not necessary, but only if best practices in software engineering are employed. These best practices have been explored and documented in detail with regard to object-oriented systems, which suggests that it is an appropriate paradigm to employ in the development of future imaging systems. This work examines these best practices, in the form of patterns and design principles, with reference to imaging systems and within the context of the Java imaging APIs.
TL;DR: This paper presents the linear array processors with multiple access modes memory system (LAPMAMM), an efficient monodimensional parallel architecture for real-time image processing that is composed of n processors and n2 memory modules.
Abstract: This paper presents the linear array processors with multiple access modes memory system (LAPMAMM), an efficient monodimensional parallel architecture for real-time image processing. This architecture is composed of n processors and n2 memory modules. These memory modules have multiple access modes: RAM, FIFO, normal CAM and interactive CAM modes. They are associated to a linear array of VLIW processors which are interconnected using a simple tree network that ensures an O(log(n)) data propagation time. The practical working of the architecture is explained using the example of a labeling algorithm developed for LAPMAMM. A hardware simulation of a LAPMAMM prototype has been carried out to test its performance in low and intermediate level image processing. The simulation results of the VHDL model are presented.