Two stages of curve detection suggest two styles of visual computation
Neural Computation
Nonlinear Image Filtering with Edge and Corner Enhancement
IEEE Transactions on Pattern Analysis and Machine Intelligence
SUSAN—A New Approach to Low Level Image Processing
International Journal of Computer Vision
Robust Image Corner Detection Through Curvature Scale Space
IEEE Transactions on Pattern Analysis and Machine Intelligence
Feature Detection with Automatic Scale Selection
International Journal of Computer Vision
Neural Mechanisms for Representing Surface and Contour Features
Emergent Neural Computational Architectures Based on Neuroscience - Towards Neuroscience-Inspired Computing
Neural mechanisms for the robust representation of junctions
Neural Computation
| Hi-index | 0.00 |
Junctions provide important cues in various perceptual tasks, such as the determination of occlusion relationship for figure-ground separation, transparency perception, and object recognition, among others. In computer vision, junctions are used in a number of tasks like point matching for image tracking or correspondence analysis. We propose a biologically motivated approach to junction detection. The core component is a model of V1 based on biological mechanisms of colinear long-range integration and recurrent interaction. The model V1 interactions generate a robust, coherent representation of contours. Junctions are then implicitly characterized by high activity for multiple orientations within a cortical hypercolumn. A local measure of circular variance is used to extract junction points from this distributed representation. We show for a number of generic junction configurations and various artificial and natural images that junctions can be accurately and robustly detected. In a first set of simulations, we compare the detected junctions based on recurrent long-range responses to junction responses as obtained for a purely feedforward model of complex cells. We show that localization accuracy and positive correctness is improved by recurrent long-range interaction. In a second set of simulations, we compare the new scheme with two widely used junction detection schemes in computer vision, based on Gaussian curvature and the structure tensor. Receiver operator characteristic (ROC) analysis is used for a threshold-free evaluation of the different approaches. We show for both artificial and natural images that the new approach performs superior to the standard schemes. Overall we propose that nonlocal interactions as realized by long-range interactions within V1 play an important role for the detection of higher order features such as corners and junctions.