An improved model for surround suppression by steerable filters and multilevel inhibition with application to contour detection

Abstract

Psychophysical and neurophysiological evidence about the human visual system shows the existence of a mechanism, called surround suppression, which inhibits the response of an edge in the presence of other similar edges in the surroundings. A simple computational model of this phenomenon has been previously proposed by us, by introducing an inhibition term that is supposed to be high on texture and low on isolated edges. While such an approach leads to better discrimination between object contours and texture edges w.r.t. methods based on the sole gradient magnitude, it has two drawbacks: first, a phenomenon called self-inhibition occurs, so that the inhibition term is quite high on isolated contours too; previous attempts to overcome self-inhibition result in slow and inelegant algorithms. Second, an input parameter called ''inhibition level'' needs to be introduced, whose value is left to heuristics. The contribution of this paper is twofold: on one hand, we propose a new model for the inhibition term, based on the theory of steerable filters, to reduce self-inhibition. On the other hand, we introduce a simple method to combine the binary edge maps obtained by different inhibition levels, so that the inhibition level is no longer specified by the user. The proposed approach is validated by a broad range of experimental results.

Image Segmentation Using a Sparse Coding Model of Cortical Area V1

“…For this reason, numerous neural network approaches have been proposed that are inspired by the physiology of the primary visual cortex (Ben-Shahar and Zucker, 2004;Grigorescu et al, 2003Grigorescu et al, , 2004Hansen and Neumann, 2008;Huang et al, 2009;Li, 1998;Mundhenk and Itti, 2005;Papari et al, 2007;Papari and Petkov, 2011;Petkov and Westenberg, 2003;Tang et al, 2007a,b;Ursino and La Cara, 2004;Vonikakis et al, 2006;Zeng et al, 2011a,b). A neuron in such a model has a classical receptive field (cRF), often defined using a Gabor function, that receives input from the image.…”

Detection of incomplete enclosures of rectangular shape in remotely sensed images

“…This technique extracts linear features, while suppressing texture elements of cluttered background. We also experimented with other approaches [20,10,28], but these are either not sensitive enough to extract faint edges of enclosures, or generate lots of clutter edges depending on the parameters used. The parameterless line segment detector [11], which is known to provide robust results for a large range of images, misses faint edges of enclosures.…”

Discriminative Generative Contour Detection

“…In order to remove small edges in highly textured regions, we use the inhibition term that suppress the response on the texture regions based on steerable filters [14]. It is computed by the convolution of the Gaussian gradient magnitude with the inhibition term (IT):…”

“…It can be shown that the difference of magnitude of Gaussian gradient and the inhibition term is the difference of Gaussian (DoG) feature [14] which provides useful edge information.…”

“…The global probability of boundary (gPb) detector [3,12] combines multiple local cues in a probabilistic framework based on spectral clustering, which shows state-of-the-art performance on the Berkeley segmentation data Set (BSDS500). Most recently, Papari et al [14] use steerable filters to construct a model with multilevel inhibition terms to remove spurious edges in textured regions for contour detection.…”