Noga Levy

The SVM-Minus Similarity Score for Video Face Recognition

Challenge, but also an opportunity to eliminate spurious similarities. Luckily, a major source of confusion in visual similarity of faces is the 3D head orientation, for which image analysis tools provide an accurate estimation. The method we propose belongs to a family of classifier-based similarity scores. We present an effective way to discount pose induced similarities within such a framework, which is based on a newly introduced classifier called SVM-minus. The presented method is shown to outperform existing techniques on the most challenging and realistic publicly available video face recognition benchmark, both by itself, and in concert with other methods.

Classification of Artistic Styles Using Binarized Features Derived from a Deep Neural Network

With the vast expansion of digital contemporary painting collections, automatic theme stylization has grown in demand in both academic and commercial fields. The recent interest in deep neural networks has provided powerful visual features that achieve state-of-the-art results in various visual classification tasks. In this work, we examine the perceptiveness of these features in identifying artistic styles in paintings, and suggest a compact binary representation of the paintings. Combined with the PiCoDes descriptors, these features show excellent classification results on a large scale collection of paintings.

Male preference for sexual signalling over crypsis is associated with alternative mating tactics

Changing body colour in animals generally reflects a conflict between two selection pressures, camouflage and social signalling. Chameleons are among the few organisms that resolve this conflict by rapid and temporary change in body colour for both background matching and social display. Male common chameleons, Chamaeleo chamaeleon, employ two alternative mating tactics, dominants and subordinates, both of which are associated with long-term body colour patterns and instantaneous colour displays during social encounters. Hence, males present a good model in which to study the influence of mating tactic on the decision of whether to remain cryptic or to signal. We exposed individuals to two conflicting external stimuli: background manipulations, which challenge camouflage, and the presence of a female, which stimulates sexual signalling. No individuals of either mating tactic responded to background manipulation except when the shift was from green to brown background or vice versa. Ambient temperatures affected colour matching but not sexual signalling, while body temperature affected neither. Males ignored the background colour and prioritized being distinctive when encountering females. As such, males were more likely to engage in sexual signalling than crypsis. Subordinate sneakers signalled females more frequently than the dominant, female-guarding males, suggesting that sneakers rapidly signal females their intentions when the dominant is out of range. Conversely, dominant males may gain little by frequent signalling to the females they guard, while possibly gaining more by diverting this energy towards mate guarding. Our results suggest that specific male mating tactics strongly influence the decision to use crypsis or sexual signalling.

Alternative Mating Tactics in Male Chameleons (Chamaeleo chamaeleon) Are Evident in Both Long-Term Body Color and Short-Term Courtship Pattern.

Alternative mating tactics in males of various taxa are associated with body color, body size, and social status. Chameleons are known for their ability to change body color following immediate environmental or social stimuli. In this study, we examined whether the differential appearance of male common chameleon during the breeding season is indeed an expression of alternative mating tactics. We documented body color of males and used computer vision techniques to classify images of individuals into discrete color patterns associated with seasons, individual characteristics, and social contexts. Our findings revealed no differences in body color and color patterns among males during the non-breeding season. However, during the breeding season males appeared in several color displays, which reflected body size, social status, and behavioral patterns. Furthermore, smaller and younger males resembled the appearance of small females. Consequently, we suggest that long-term color change in males during the breeding season reflects male alternative mating tactics. Upon encounter with a receptive female, males rapidly alter their appearance to that of a specific brief courtship display, which reflects their social status. The females, however, copulated indiscriminately in respect to male color patterns. Thus, we suggest that the differential color patterns displayed by males during the breeding season are largely aimed at inter-male signaling.

Congruency-Based Reranking

We present a tool for re-ranking the results of a specific query by considering the (n+1) × (n+1) matrix of pairwise similarities among the elements of the set of n retrieved results and the query itself. The re-ranking thus makes use of the similarities between the various results and does not employ additional sources of information. The tool is based on graphical Bayesian models, which reinforce retrieved items strongly linked to other retrievals, and on repeated clustering to measure the stability of the obtained associations. The utility of the tool is demonstrated within the context of visual search of documents from the Cairo Genizah and for retrieval of paintings by the same artist and in the same style.

Minimal correlation classification

When the description of the visual data is rich and consists of many features, a classification based on a single model can often be enhanced using an ensemble of models. We suggest a new ensemble learning method that encourages the base classifiers to learn different aspects of the data. Initially, a binary classification algorithm such as Support Vector Machine is applied and its confidence values on the training set are considered. Following the idea that ensemble methods work best when the classification errors of the base classifiers are not related, we serially learn additional classifiers whose output confidences on the training examples are minimally correlated. Finally, these uncorrelated classifiers are assembled using the GentleBoost algorithm. Presented experiments in various visual recognition domains demonstrate the effectiveness of the method.

active congruency-Based reranking

We present a tool for re-ranking the results of a specific query by considering the matrix of pairwise similarities among the elements of the set of retrieved results and the query itself. The re-ranking thus makes use of the similarities between the various results and does not employ additional sources of information. The tool is based on graphical Bayesian models, which reinforce retrieved items strongly linked to other retrievals, and on repeated clustering to measure the stability of the obtained associations. To this, we add an active relevance-based re-ranking process in order to leverage true matches, which have very low similarity to the query. The utility of the tool is demonstrated within the context of a visual search of documents from the Cairo Genizah. It is also demonstrated in a completely different domain or retrieving, given an input image of a painting, other related paintings.

Gait-Based Person Identification Using Motion Interchange Patterns

Understanding human motion in unconstrained 2D videos has been a central theme in Computer Vision research, and over the years many attempts have been made to design effective representations of video content. In this paper, we apply to gait recognition the Motion Interchange Patterns (MIP) framework, a 3D extension of the LBP descriptors to videos that was successfully employed in action recognition. This effective framework encodes motion by capturing local changes in motion directions. Our scheme does not rely on silhouettes commonly used in gait recognition, and benefits from the capability of MIP encoding to model real world videos. We empirically demonstrate the effectiveness of this modeling of human motion on several challenging gait recognition datasets.