A Design of Zynq-based Medical Image Edge Detection Accelerator

Abstract

Edge detection technology plays an important role in medical image processing. Sobel operator edge detection is one of the commonly used edge detection operators. At present, most of the solutions using Sobel operator for edge detection of medical images are based on CPU and GPU. Processing speed can become a serious problem as image data increases. The acceleration effect of FPGA on edge detection is quite significant. However, the traditional Sobel edge detection scheme based on FPGA is developed by hardware description language, which has high requirements for developers and is very unfavorable to debugging. Using the Zynq series of C/C++ programming for acceleration can perfectly solve the above problems. However, the current Zynq-based Sobel operator edge detection research, only horizontal edge and vertical edge detection. In order to extract more edge details from different angles, we proposed an improved Sobel operator based on Zynq to detect edges. The performance of the proposed improved Sobel algorithm and the conventional Sobel algorithm on CPU and Zynq platform is compared and evaluated in detail. Experimental results show that the proposed scheme can extract more edge details and achieve satisfactory acceleration effect.