Detecting Irregular Curvilinear Structures in Gray Scale and Color Imagery Using Multi-directional Oriented Flux

Abstract

We propose a new approach to detecting irregular curvilinear structures in noisy image stacks. In contrast to earlier approaches that rely on circular models of the crosssections, ours allows for the arbitrarily-shaped ones that are prevalent in biological imagery. This is achieved by maximizing the image gradient flux along multiple directions and radii, instead of only two with a unique radius as is usually done. This yields a more complex optimization problem for which we propose a computationally efficient solution.We demonstrate the effectiveness of our approach on a wide range of challenging gray scale and color datasets and show that it outperforms existing techniques, especially on very irregular structures.

Multiscale Centerline Detection

“…For example, in [27], an Oriented Flux Antisymmetric (OFA) term was added and has proved effective. There has been less work on improving OOF's performance on truly irregular structures, except for the very recent approach of [53] that attempts to maximize the image gradient flux along multiple radii in different directions instead of only one as in [26].…”

“…Even if care is taken to add computational machinery to handle irregular structures [43], [53], the performance of handdesigned filters tends to suffer in severe cases such as the one depicted by Fig. 1.…”

“…In this way, performing non-maximum suppression on their output yields both centerline locations and corresponding radii. We will show that, on very irregular structures, it outperforms the powerful OOF approach with and without anti-symmetry term [26], [27], which is widely acknowledged as one of the best among those relying on hand-designed filters; a very recent extension of it [53] designed to improve its performance on irregular structures; and a similarly recent classificationbased method [5]. We will also evaluate the ability of our method to trace the linear structures.…”

“…We will also evaluate the ability of our method to trace the linear structures. In particular, we will demonstrate that feeding our output as input to a complete tracing algorithm [54] instead of that of [53] increases final performance.…”

“…More precisely, we train several regressors to return distances to the closest centerline Comparison of the responses of our method against a recent model based approach [53] and a classification based one [5]. Bottom row: Centerlines detected after performing Non-Maximum Suppression on the response images.…”