A self-organizing neural system for learning to recognize textured scenes

Abstract

A self-organizing ARTEX model is developed to categorize and classify textured image regions. ARTEX specializes the FACADE model of how the visual cortex sees, and the ART model of how temporal and prefrontal cortices interact with the hippocampal system to learn visual recognition categories and their names. FACADE processing generates a vector of boundary and surface properties, notably texture and brightness properties, by utilizing multi-scale filtering, competition, and diffusive filling-in. Its context-sensitive local measures of textured scenes can be used to recognize scenic properties that gradually change across space, as well as abrupt texture boundaries. ART incrementally learns recognition categories that classify FACADE output vectors, class names of these categories, and their probabilities. Top-down expectations within ART encode learned prototypes that pay attention to expected visual features. When novel visual information creates a poor match with the best existing category prototype, a memory search selects a new category with which classify the novel data. ARTEX is compared with psychophysical data, and is bench marked on classification of natural textures and synthetic aperture radar images. It outperforms state-of-the-art systems that use rule-based, backpropagation, and K-nearest neighbor classifiers.

Self-Organization of Topographic Mixture Networks Using Attentional Feedback

“…On the whole, GAM's performance did not improve with further training. As reported in Grossberg and Williamson (1998), these results arc Buperior to those obtained on a similar texture classification task by an alternative image classification architecture that used rule-based, multilayer perceptron, or knearest neighbor classifiers. GAM also outperformed this alternative architecture when they wme both evaluated on an identical task involving the classification of 10 natural textures.…”

“…In order to evaluate the eJTeet of these changeB, 'I'AM was compared to GAM on natural texture classification benchmarks, on which GAM obtajned good results (Grossberg and Williamson, 1998). The dataset was produced by a biologically-motivated image processing system which extracted, from each 8x8 pixel region of an input image, 16 oriented contrast features (four orientations a.nd four spatial scales) a.s well as a. single brightness feature.…”

A computational feature binding model of human texture perception

“…"Back pocket models" based on the former were presented by (Fogel & Sagi, 1989;Malik & Perona, 1990;Landy & Bergen, 1991;Manjunath & Chellappa, 1993). The models of (Jain & Farrokhina, 1991;Grossberg & Williamson, 1999;Hofmann, Puzicha, & Buhmann, 1998) are using region based mechanisms to model texture segregation phenomena. However, psychophysical studies have revealed that in human texture perception both mechanisms are present: Some experiments (Nothdurft, 1985) have demonstrated that for certain stimuli edge based mechanisms are utilized, whereas other studies (Wolfson & Landy, 1998) showed that region based mechanisms are necessary to explain human texture segregation.…”

“…Although the texture segmentation performance of (Hofmann et al, 1998) is very good, this heuristic and the annealing framework itself are biologically not very plausible. A more promising approach is that from Grossberg and Williamson (Grossberg & Williamson, 1999). Their ARTEX model incorporates a self organizing network which gives remarkable classification results on multiple scale orientational contrast texture features.…”

The laminar architecture of visual cortex and image processing technology

“…Even without self-organization, such circuits have already yielded the most effective self-equilibrating boundary and surface representation algorithm that we have ever used for processing Synthetic Aperture Radar images [4]. In addition, simplified versions of these boundary mechanisms have been used to develop an architecture for self-organizing recognition categories in response to natural textures and textured Synthetic Aperture Radar images [ 43].…”