Multifeature-Based Surround Inhibition Improves Contour Detection in Natural Images

Abstract

To effectively perform visual tasks like detecting contours, the visual system normally needs to integrate multiple visual features. Sufficient physiological studies have revealed that for a large number of neurons in the primary visual cortex (V1) of monkeys and cats, neuronal responses elicited by the stimuli placed within the classical receptive field (CRF) are substantially modulated, normally inhibited, when difference exists between the CRF and its surround, namely, non-CRF, for various local features. The exquisite sensitivity of V1 neurons to the center-surround stimulus configuration is thought to serve important perceptual functions, including contour detection. In this paper, we propose a biologically motivated model to improve the performance of perceptually salient contour detection. The main contribution is the multifeature-based center-surround framework, in which the surround inhibition weights of individual features, including orientation, luminance, and luminance contrast, are combined according to a scale-guided strategy, and the combined weights are then used to modulate the final surround inhibition of the neurons. The performance was compared with that of single-cue-based models and other existing methods (especially other biologically motivated ones). The results show that combining multiple cues can substantially improve the performance of contour detection compared with the models using single cue. In general, luminance and luminance contrast contribute much more than orientation to the specific task of contour extraction, at least in gray-scale natural images.

Hierarchical linear dynamical systems for unsupervised musical note recognition

“…So these are somber predictions, but on the positive side, the states are infinite functional spaces, as is the output of the observation model, which can be very rich. We have already demonstrated that HLDS dynamics approximate well attractors that are an interesting property of nonlinear models and useful in classification (the Hopfield network[12]), and we will also show that linear inference is capable of approximating surround inhibition[27], a property of nonlinear models. Moreover, learning is extremely efficient because HLDS preserves the recursive update of the Kalman filter.Instead of discarding summarily HLDS for machine learning applications, we should ask: what type of interesting problems can be learned with linearly separable inference in functional spaces?…”

“…Computational models that mimic surround suppression (surround inhibition) are used in a variety of tasks such as motion detection and noise suppression [46], contour detection [27], speech processing [47]. How the auditory cortex maps the auditory space is a key factor in the surround suppression.…”

Contrast-dependent surround suppression models for contour detection

“…Ursino et al [16] combined feed-forward and feedback inhibition mechanisms to present a model of contour extraction, in which the feed-forward input plays a major role to suppress non-optimal stimuli and to ensure contrast invariance, and long-range feedback inhibition is essential to suppress noise. Li and colleagues [17][18][19] have developed biologically-inspired contour perception models over many years, especially in non-CRF inhibition modeling. They applied multi-resolution techniques to surround suppression models in which an adaptive end inhibition is realized based on the information extracted at two different spatial scales -Gabor energy information at a coarse scale and the side inhibition information at a fine scale [17].…”

“…They applied multi-resolution techniques to surround suppression models in which an adaptive end inhibition is realized based on the information extracted at two different spatial scales -Gabor energy information at a coarse scale and the side inhibition information at a fine scale [17]. They have also recently proposed a multiple-cue inhibition method in which the multi-feature combined weights are used to modulate the final inhibitory responses of the neurons [18]. These models can dramatically reduce trivial edges from texture regions and thus improve the performance of contour detection in comparison to traditional local operators.…”

A biologically inspired framework for contour detection

“…Another way to model the response of LGN cells has also been recently proposed in [31], in which the authors propose a computational model of a simple cell, which combines the responses of model LGN cells with center-surround receptive field and that outperforms the traditional Gabor function model in contour detection. Finally, we can also mention the sparse coding model of cortical area proposed in [32], the combined use of retina and cerebral cortex modeling presented in [33] for the development of low-level image processing tasks such as motion analysis and video data structuring and the multifeature-based center-surround framework proposed in [34].…”