Florin Cutzu

The selective tuning model of attention: psychophysical evidence for a suppressive annulus around an attended item

The selective tuning model [Artif. Intell. 78 (1995) 507] is a neurobiologically plausible neural network model of visual attention. One of its key predictions is that to simultaneously solve the problems of convergence of neural input and selection of attended items, the portions of the visual neural network that process an attended stimulus must be surrounded by inhibition. To test this hypothesis, we mapped the attentional field around an attended location in a matching task where the subjects attention was directed to a cued target while the distance of a probe item to the target was varied systematically. The main result was that accuracy increased with inter-target separation. The observed pattern of variation of accuracy with distance provided strong evidence in favor of the critical prediction of the model that attention is actively inhibited in the immediate vicinity of an attended location.

Estimating the photorealism of images: distinguishing paintings from photographs

Automatic classification of an image as a photograph of a real-scene or as a painting is potentially useful for image retrieval and Web site filtering applications. The main contribution of the paper is the proposition of several features derived from the color, edge, and gray-scale-texture information of the image that effectively discriminate paintings from photographs. For example, we found that paintings contain significantly more pure-color edges, and that certain gray-scale-texture measurements (mean and variance of Gabor filters) are larger for photographs. Using a large set of images (12000) collected from different Web sites, the proposed features exhibit very promising classification performance (over 90%). A comparative analysis of the automatic classification results and psychophysical data is reported, suggesting that the proposed automatic classifier estimates the perceptual photorealism of a given picture.

Distinguishing paintings from photographs

We addressed the problem of automatically differentiating photographs of real scenes from photographs of paintings. We found that photographs differ from paintings in their color, edge, and texture properties. Based on these features, we trained and tested a classifier on a database of 6000 paintings and 6000 photographs. Using single features results in ~70-80% correct discrimination performance, whereas a classifier using multiple features exceeds 90% correct discrimination.

Polychotomous classification with pairwise classifiers: a new voting principle

A new principle for performing polychotomous classification with pairwise classifiers is introduced: if pairwise classifier Nij, trained to discriminate between classes i and j, responds i for an input x from an unknown class (not necessarily i or j), one can at best conclude that x ∉ j. Thus, the output of pairwise classifier Nij can be interpreted as a vote against the losing class j, and not, as existing methods propose, as a vote for the winning class i. Both a discrete and a continuous classification model derived from this principle are introduced.

Differences of edge properties in photographs and paintings

We compare the properties of intensity and color edges in photographs of real scenes and paintings. We demonstrate that paintings contain significantly more color-only edges, whereas the amount of intensity-only edges does not differ significantly between the two classes. In addition, color edge strength is significantly higher for paintings. The differences between paintings and photographs are more accentuated when high-resolution, losslessly compressed images are used. These distinguishing features can be used for the automatic differentiation between the two classes of images.

Hierarchical, Generic to Specific Multi-class Object Recognition

Early vision systems could perform specific object recognition reasonably well but did not fare as well on identifying natural object classes. Recent advances have led to systems that can learn a representation for different object classes and achieve good generic object class recognition. However, these systems are generally unable to perform the fine distinctions required for specific object identification. It seems that the two approaches are in contrast with each other. We propose a system that addresses the problems of generic class recognition as well as specific object recognition in the same framework. Our system also possesses the property of graceful degradation, i.e., if it is unable to recognize an object as a dog, it recognizes it at least as a quadruped and so on. We automatically learn a hierarchy of the classes from the training data, progressing from the most generic class labels to the most specific object labels. This hierarchy is used during recognition. One important benefit of the hierarchical organization of the classes is that the number of comparisons performed for every input does not increase linearly with the number of classes added

How to do multi-way classification with two-way classifiers

A new principle for performing polychotomous classification with pairwise classifiers is introduced: if pairwise classifier Nij, trained to discriminate between classes i and j, responds i for an input x from an unknown class (not necessarily i or j), one can at best conclude that x ∉ j. Thus, the output of pairwise classifier Nij can be interpreted as a vote against the losing class j, and not, as existing methods propose, as a vote for the winning class i. Both a discrete and a continuous classification model derived from this principle are introduced.

Computing Disparity on Demand: Disparity based Classification using Error-Tolerant Decision Tree Ensembles.

Most range-based recognition systems require the calculation of a full disparity map at adequate resolutions prior to the recognition step. There also exist range-based systems that only require the computation of a sparse disparity map. We introduce a 3D shape classification method in which the disparity calculation is guided by the needs of the classification process. The method uses decision trees for shape classification and calculates the disparity only at certain locations in the image, as required by the tree structure. The calculation is very efficient as only the minimum number of disparity values are calculated. To render the classification robust, we use an ensemble of trees. The proposed ensemble method is different from the currently known ensemble methods and makes the classification system more robust to errors in the disparity calculation. The method was applied to a real world problem and good classification results were obtained.

Using Multiple Views To Resolve Human Body Tracking Ambiguities

This paper outlines the theoretical background and presents a new approach to human body tracking with monocular static camera. A novel “viewbased representation” is introduced at the feature extraction stage. We show that ambiguities in correspondence, such as the ones that occur as the result of occlusion, can be resolved by using this approach. In particular, we store color information for each object in a vector of views, where the number of elements is determined online, using unsupervised clustering followed by the cluster validity assessment. Based on this representation a tracking system was developed. The prelimiary results presented show the discriminative potential of the proposed system.