A new real time edge linking algorithm and its VLSI implementation

Edges produced by existing edge detection algorithms often contain discontinuities. Edge linking as a post-processing step is an important step for computer vision and pattern recognition. We present a real-time algorithm and its VLSI implementation for linking broken edges. First, all broken edge points inside a 12/spl times/12 moving window are identified. The 12/spl times/12 window scans the input gray level edge map converting it into three levels of intensities using two threshold values. Decisions of linking the stronger edge break points are made based on their directions and/or with the guidance of the weak edge lines. The proposed VLSI architecture is capable of running the proposed edge linking algorithm in real-time outputting one pixel of the linking edge map per clock cycle with a latency of 11n+12 clock cycles, where n is the number of pixel columns in the image.