Unsupervised edge detection and noise detection from a single image

Edge detection is one of the oldest image processing areas that are still active. An important current area of study involves development of unsupervised edge detection algorithms. In this work a paradigm of unsupervised edge detection is proposed that is based on the computational edge detection approach introduced by Canny. It is a simple and computationally cheap technique that achieves non-trivial results. Additionally as a byproduct it generates information about the content and severity of noise in the image. The proposed technique uses a fast edge detector to generate the initial edge mask and subsequently optimizes that by studying the behavior of a proposed details estimator. The study of the same estimator also offers insight about the noise characteristics of the image.

► We present a simple paradigm for unsupervised edge detection.
► A simple estimator is studied in conjunction with a modified non-maximal suppression algorithm.
► The study of the same estimator indicates presence and extent of noise in the image.