V. K. Prasanna and C. L. Wang. Image feature extraction on connection machine CM-5. In Proceedings of the Image Understanding Workshop, volume 1, pages 595--602, 1994.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....3] the Hough transform [4] energy based methods [13] and the use of extra information such as colour [12] and temporal sequences [5] Most of these have been aimed at removing or avoiding gaps in the edges. Others have sought faster edge processing by implementing parallel processing approaches [2, 9]. Here we aim for speed while using conventional sequential execution hardware. The methods used result in linked edges that may not Gamma1 1 Gamma1 1 Gamma1 1 I x 1 1 1 Gamma1 Gamma1 Gamma1 I y Figure 1. Masks for the Prewitt edge detector. be of the highest quality possible, ....

V. K. Prasanna and C. L. Wang. Image feature extraction on connection machine CM-5. In Proceedings of the Image Understanding Workshop, volume 1, pages 595--602, 1994.

....approximate each contour into line segments. Parallelizing the contour pixels detection phase on a coarse grain machine such as SP 2, can be done easily. After exchanging the boundary information, the operations in a processing node can be done independently of operations in other processing nodes [Prasanna and Wang, 1994]. However, in the linear approximation phase, the run time data dependencies make the design of a parallel algorithm and its implementation to obtain large speed ups to be a nontrivial task. The irregularity causes some processing nodes to be idle while others are active. In this paper, we only ....

....into the system buffer. Thus, the execution time of the blocking send command depends on the message size. Since both these send commands do not guarantee that the message has been delivered to the receiver, we also measured the round trip communication time by a ping pong operation. As in [Prasanna and Wang, 1994], the roundtrip communication time can be modeled as T d m d time, where T d denote the startup time for sending a message and d denote the transmission rate (seconds per bytes) for data communication; T d = 46 sec and d = 0:035 sec=byte. Therefore, if the size of message is smaller than ....

V. Prasanna and C. Wang, "Image Feature Extraction on Connection Machine CM-5," Image Understanding Workshop, pp. 595-602, 1994.

....of the code to various platforms. We believe modeling the features of parallel machines, designing data partitioning techniques, and mapping the computations to balance the load using explicit message passing is a feasible approach to solve vision problems on distributed memory parallel machines [1, 6, 10, 11]. Our algorithm can be ported to other models [14] including for example, the shared memory model discussed in [2] Acknowledgment This research was supported in part by NSF under grant CCR 9317301 and in part by ARPA under grant F49620 93 1 0620. The access to T3D was provided by the ....

V. Prasanna and C-L Wang, "Image Feature Extraction on Connection Machine CM-5," Image Understanding Workshop, pp. 595-602, November 1994.

....approximate each contour into line segments. Parallelizing the contour pixels detection phase on a coarse grain machine such as SP 2, can be done easily. After exchanging the boundary information, the operations in a processing node can be done independently of operations in other processing nodes [14]. However, in the linear approximation phase, the runtime data dependencies make the design of a parallel algorithm and its implementation to obtain large speed ups to be a nontrivial task. The irregularity causes some processing nodes to be idle while others are active. We develop an asynchronous ....

....and the experimental results are shown in Section 5. Finally, concluding remarks are made in Section 6. 2 Linear Feature Extraction Linear Feature Extraction consists of ContourPixels Detection and Linear Approximation phases. In the following, we briefly describe these algorithms [13, 14, 15]. The input to the contour pixels detection is a 2 D image array of pixels (grey levels) The output is a same sized array with directed contour pixels embedded in the array. The contour pixels detection consists of edge detection, thinning, and linking operations. These are window operations, in ....

[Article contains additional citation context not shown here]

V. Prasanna and C. Wang, "Image Feature Extraction on Connection Machine CM-5," Image Understanding Workshop, pp. 595-602, 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