Content area
Full text
Abstract - Edge detection is a considerably important factor in image or video processing. Detection of edges plays a significant role in image segmentation, data compression, well matching, and image reconstruction. Among several edge detection approaches we focus on Sobel edge detection using contract-time anytime algorithm in CUDA. To reduce the computational complexity we implemented our proposed edge detection method using CUDA. In the experimental setup we have used NVIDIA GTX 550Ti GPU along with AMD FX8150 Processor and 8 GB RAM. Finally, we measure speedup using 3 steps of contract-time anytime of our proposed parallel implementation model. Comparing with conventional serial CPU based edge detection we have experienced maximum 4X speedup of proposed implementation for 16 block dimension.
Keywords: Edge detection; CUDA; Anytime algorithm; Parallel Computing; Sobel; NVIDIA
(ProQuest: ... denotes formulae omitted.)
1 Introduction
Edge detection from a color image is a very important and basically critical area in low level image processing. For performing high speed industrialized application based on image processing, edge detection is a mandatory thing to enhance work rate as well as accuracy. A number of researchers works on several edge detection algorithms and they give different responses and details to the different input images [1-7]. Edge detection quality has a great impact on realization of complex automated computer/machine vision systems [1]. Among them, the Sobel edge detection algorithm is much more popular than simple gradient operators due to its property to counteract the noise sensitivity and easier implementation process [2]. While using Sobel operator for GPU takes much less time than CPU. Again, the use of Interruption-Algorithm for image processing much less time efficient [3]. Moreover, in case of canny edge detection in GPU time process seems efficient but not enough for real time [4]. The use of anytime algorithm for GPU architecture makes 8]. In addition, anytime algorithm seems much efficient when it is used for observing different tasks [6]. Interruptible Anytime Algorithm for image processing is much faster than normal image processing algorithms and also gives the privilege of getting output in different stage of time [7]. That is why, we are choosing contract-time anytime algorithm in co-ordination with Sobel operator for proposed parallel implementation.
The rest of this paper is organized as follows:...