Distributed Canny Edge Detector: Algorithm and FPGA Implementation of NMS Block

Abstract

The Canny edge detector is a widely used edge detection algorithms because of its superior performance. It was developed by John F. Canny in 1986. The computation is based on frame-level statistics which results in higher latency. To overcome this, block-level computation is used rather than frame-level, but applying directly Canny edge algorithm to blocks results in excessive edges in smooth regions and loss of significant edges in the highly detailed region. Hence Distributed Canny Edge Algorithm is adapted, which computes the edge detection thresholds based on the type of block and local distribution of the gradients of the pixels in the image. This algorithm also uses a non-uniform gradient magnitude histogram for computing block-based hysteresis thresholds. The block-based algorithm significantly reduces latency and can also be easily integrable with other block-based image codecs.