Efficient image edge detection method using least squares and elbow technique.

We propose a new edge detection method using the least squares and a novel elbow technique. The method approximates the local intensity function with a 2D quadratic function using the least squares and computes its gradients to obtain edge response values. We introduce an elbow technique that provides a suitable threshold value for single-threshold edge detection from the graph of percentile positions versus edge response values. We also combine the elbow technique with non-maximum suppression to provide high and low thresholds for double-threshold edge detection, enabling edge thinning. The results show that the proposed single-threshold and double-threshold edge detections can capture all edge structures. Moreover, the double-threshold detection can produce thinner edges.