On the application of massively parallel SIMD tree machines to certain intermediate-level vision tasks

Hussein A.H. Ibrahim, John R. Kender, David Elliot Shaw
1986 Computer Vision Graphics and Image Processing  
In thIS paper, we examine the implementation of two middle-level Image understanding tasks on fine-grained tree-structured Sl1m machines, which have highly effIclent v1.SI implementations. We first present one such massively parallel machine called NON-VON, and summarize the cost/performance tradeoffs of such machines for vision tasks. \Ve follow with a more detailed description of the P;ON-VO~ architecture (a prototype of which has been operational since January, 1985), and of the high-level
more » ... rallel language in whIch our algorithms have been WrItten and simulated The heart of thE' paper consists of the deSCrIption and analysls of algorIthms for a representative Hough transform, and of an algorIthm for the interpretation of mOVlf:g light displays. Novel algorithmic techniques are motivated and described, and sim ulation timings are presented and discussed. \Ve conclude that it lS possible to exploit the available masslve parallelism whIle avoiding many of the com m unication bottlenecks common at this level of lmage understanding, by carefully and inexpenslvely duplicating data and/or control Information, and by delaYIng or aVOIding the reporting of intermediate results. Index terms VislOn hardware, Hough Transform, moving light displays, parallel processing Both algOrithms have been Simulated using a functional Simulator runmng on a VA .. X 11/750 augmented with a Grinnell Image processor. Other Image understanding tasks (not discussed in thiS paper) for which NON-VON algOrithms have been developed and simulated include image correlation, histogrammmg, thresholding, connected component labeling, and the computation of the area, perimeter, center of gravity, eccentncity, and Euler number of connected components [101. Based on simulation results, NON-VON's performance has been compared with that of other highly parallel architectures for image analysis Certain algorithms have been shown to execute faster on NON-VON than on other highly parallel machines having a similar cost. cost/performance advantages were seen to derive from These performance and 1. The effective use of an unusually high degree of parallelism, made possIble by the machine's very fine granularity. 2. The natural mapping of hierarchical and multi-resolutIon techniques developed by other researchers onto NON-VON's tree structured topology 3. The extensive use of content-addressable matching and other asSOCIatIVe processing technIques. 4. The use of the tree to perform algebraically associative operatIons such as addition in time logarithmic in the number of pixels. 5. The slmphcity and cost-effectiveness with which tree-structured machines can be Implemented using VLSI technology.
doi:10.1016/0734-189x(86)90038-1 fatcat:rp2l6hy2lzey7cyjlzsua3zlmq