Ronen Basri: Research

Edge detection and perceptual grouping

 

See my talk on this subject in YouTube

 

A fundamental question for edge detection is how faint an edge can be and still be detected. Our work in this area offers a formalism to study this question and introduces a family of hierarchical edge detection algorithms designed to detect faint edges in noisy images. In our formalism we view edge detection as a search in a space of feasible curves, and derive expressions to characterize the behavior of the optimal detection threshold as a function of curve length and the combinatorics of the search space. We developed two algorithms that efficiently search for edges  — the first algorithm applies straight filters of varying length and orientation. The second algorithm searches through a very large set of curves by hierarchically constructing difference filters that match the curves traced by the sought edges. We further offer a method for fiber detection by applyung a diffusion process to the detect edges. Our approach manages to detect edges in very low SNRs, but is also useful for natural images, see examples below.

A blowup of a noisy Image (top) acquired by electron microscope and the intensity profiles along the two sides of an edge (bottom).

Our probabilistic theory of edge detection by multiscale filters is introduced in
                Sharon Alpert, Meirav Galun, Boaz Nadler, and Ronen Basri, “Detecting faint curved edges in noisy images,”                 European Conf. on Computer Vision (ECCV-10), Crete, Greece, 2010.

Our straight line edge detector is described in
                Meirav Galun, Ronen Basri, and Achi Brandt, “Multiscale edge detection and fiber enhancement using differences of                  oriented means,” 11^th IEEE Int. Conf. Computer Vision (ICCV-07), Rio de Janeiro, Brazil, 2007.

Approximations to the curve of least energy and a multiscale approach for curve completion was published in
                Eitan Sharon, Achi Brandt, and Ronen Basri, “Completion energies and scale,” IEEE Transactions on Pattern Analysis and                 Machine Intelligence, 22(10): 1117-1131, 2000.

Our analysis of Shashua and Ullman’s Saliency Network has appeared in
                Tao D. Alter and Ronen Basri, “Extracting salient curves from images: an analysis of the saliency network,” International                 Journal of Computer Vision, 27(1): 51-69, 1998.

Another interesting method for grouping is described in
                Yossi Cohen and Ronen Basri, “Inferring Region Salience from Binary and Gray-Level Images,” Pattern recognition, 36(10), 2349-2362, 2003.

Position (along the edge)

Edge detection applied to electron microscopy (left set) and natural images (right set). Both sets show the original image (left) our result (middle) and results obtained with the Canny algorithm (right). Notice in particular the dense narrow stripes in the left set which are completely missed by Canny.

In previous work we also addressed problems in perceptual grouping, i.e., the task of extracting smooth long curves in images while overcoming gaps. In one paper we derived simple but accurate approximations to the curve of least energy and described an efficient multiscale algorithm for curve completion. In another paper we analyzed the ‘Saliency Network’ of Shashua and Ullman and showed difficulties inherent to the reliance on dynamic programming such as ‘leeching’ (non salient curves that are ranked highly due to their proximity to salient ones) as well as discretization problems. Code for the Saliency Network implemented for this work is given here.

What lies ahead? Perceptual grouping algorithms can potentially complement edge detection schemes to yield robust extraction of curves. Also, the difficulties in detecting edges are further aggravated in higher dimensions.

An implementation  of Shashua and Ullman’s Saliency Network written by Tao D. Alter is available online here.

The Saliency Network: from left to right, input image, saliency map and most salient curve returned by the method.