A. C. Downton. Generalised approach to parallelising image sequence coding algorithms. IEE Proceedings Part I (Vision, Image and Signal Processing), 141(6):438-445, 1994.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....are a natural structure. Soft real time systems do not generally have to meet critical deadlines though they do have to meet performance targets such as a given throughput or traversal latency. Such systems are datacentric, not control centric. Candidate applications exist in video coding [ 1 ] vision [ 2 ] and image processing [ 3 ] Recent industrial interest in networked embedded systems [ 4 ] has led us to consider the modi cations needed to cater for processing pipelines across mobile networks. APTT is built around a design pattern consisting of a pipeline of data farms each ....

A. C. Downton. Generalised approach to parallelising image sequence coding algorithms. IEE Proceedings Part I (Vision, Image and Signal Processing), 141(6):438-445, 1994.

....software tools and a generic design strategy. The approach taken in this paper is to use a generalised method of system design, pipelined processor farms (PPF) 5 ] The aim of this methodology is to introduce exibility and reduce software development costs by imposing a generic solution [ 6, 3 ] PPF is suited to the parallelisation of legacy code, or code written on workstations supporting a sequential pro3 gramming model. The algorithmic developers need not be the same persons responsible for parallelisation. In other words, the code may be unfamiliar requiring an initial exploratory ....

....which would result in a four fold increase in processing time on a workstation. The remainder of the paper is organised thus. In Section 2, the initial design stage, which involves exploration of the algorithm, is described. Since the parallelisation took place in incremental stages, Section 3 describes the rst pipeline stage and Section 4 discusses subsequent stages. Section 4 also discusses results of the paralllelisation. Finally, Section 5 summarises and draws some conclusions. 2 Analysis of the Algorithm Though parallelisation of face recognition has taken place before, a ....

[Article contains additional citation context not shown here]

A. C. Downton. Generalised approach to parallelising image sequence coding algorithms. IEE Proceedings Part I (Vision, Image and Signal Processing), 141(6):438-445, 1994.

.... frequently be combined into chunks (nomenclature adopted from the performance literature cited) For example, we have parallelized handwritten postcode recognition [ 1 ] with characters (tasks) within a postcode (UK zipcode) chunk) from a file of postcode images (job) a hybrid video encoder [ 22 ] with macro blocks (tasks) of a video row (chunk) taken from a sequence of frames (job) and 2 D frequency transforms [ 23 ] with rows (tasks) in a sub image (chunk) from a batch of images (job) PPF has been applied to low cost distributed memory machines using a message passing programming ....

....be built up in a stage by stage fashion. Once the maximum traversal latency of an asynchronous segment is found, it can be treated as an estimate for a single stage within a synchronous pipeline. The H. 261 low bit rate hybrid video encoder [ 49 ] has been parallelized by the PPF methodology [ 22 ] to enable further rapid algorithmic prototyping. Figure 3 shows a block structure that can be reduced to a linear pipeline by suitable combination of functions, Figure 4. The linear pipeline is largely synchronous due to the remaining feedback loops. A static timing analysis, Table 3, gave the ....

A. C. Downton. Generalised approach to parallelising image sequence coding algorithms. IEE Proceedings Part I (Vision, Image and Signal Processing), 141(6):438--445, 1994. 33

....Each stage of the pipeline then exploits parallelism in the most appropriate way, for example, data parallelism applied at various levels, algorithmic parallelism, or temporal multiplexing of complete data sets. This decomposition has been applied successfully to similar applications in the past [3, 6]. 3.2 Static analysis of H.263 algorithm The first step of the design model is to decompose the sequential software into pipeline stages using top down profiling data derived from sequential execution. This stage identifies the individual stages and their relative computational requirements and ....

A. C. Downton "Generalised Approach to Parallelising Image Sequence Coding Algorithms", IEE Proc. Vis. Image Signal Process., Vol. 141, No. 6, pp. 438--445, Dec. 1994.

....frequently be broken into jobs. We define the duration of a test run to be the time to complete a set of tasks. For example, we have parallelized handwritten postcode recognition [ 7 ] with characters (jobs) within a postcode (task) from a file of postcode images (run) a hybrid video encoder [ 8 ] with blocks (jobs) of a video row (task) taken from a sequence of frames (run) and 2 D frequency transforms [ 9 ] with rows (jobs) in a sub image (task) from a batch of images (run) The methods considered in this paper assume that there exists some knowledge of the service time ....

....minms is too pessimistic. Refer again to Table 1 for the timings. 5.2 Simulation results A simulation was conducted to verify the large p behaviour of the various estimators. To approach the population statistics 20,000 jobs were distributed. Experiments were made at eight processor intervals from 8 to 248 processors. 6 Equation 18 will be recognized as a variant of Riemman s zeta function. 12 Figs. 5 7 are the result, for a sample range of processor numbers, of varying the task size from 1 to 8 jobs with exponential service distribution. In regard to uniform task sizes, for small ....

[Article contains additional citation context not shown here]

A. C. Downton. Generalised approach to parallelising image sequence coding algorithms. IEE Proceedings Part I (Vision, Image and Signal Processing), 141(6):438--445, 1994.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC