A visual attention system, inspired by the behavior and the neuronal architecture of the early primate visual system, is presented. Multiscale image features are combined into a single topographical saliency map. A dynamical neural network then selects attended locations in order of decreasing saliency. The system breaks down the complex problem of scene understanding by rapidly selecting, in a computationally efficient manner, conspicuous locations to be analyzed in detail.

/pdf/a-model-of-saliency-based-visual-attention-for-rapid-scene-wv0n3yqxix.pdf

A model of saliency-based visual attention for rapid scene analysis

Probability Distributions.- Linear Models for Regression.- Linear Models for Classification.- Neural Networks.- Kernel Methods.- Sparse Kernel Machines.- Graphical Models.- Mixture Models and EM.- Approximate Inference.- Sampling Methods.- Continuous Latent Variables.- Sequential Data.- Combining Models.

Pattern Recognition and Machine Learning

Data Mining Practical Machine Learning Tools and Techniques

This paper presents a method for recognizing scene categories based on approximate global geometric correspondence. This technique works by partitioning the image into increasingly fine sub-regions and computing histograms of local features found inside each sub-region. The resulting "spatial pyramid" is a simple and computationally efficient extension of an orderless bag-of-features image representation, and it shows significantly improved performance on challenging scene categorization tasks. Specifically, our proposed method exceeds the state of the art on the Caltech-101 database and achieves high accuracy on a large database of fifteen natural scene categories. The spatial pyramid framework also offers insights into the success of several recently proposed image descriptions, including Torralbas "gist" and Lowes SIFT descriptors.

/pdf/beyond-bags-of-features-spatial-pyramid-matching-for-4aa5cbe2fz.pdf

Beyond Bags of Features: Spatial Pyramid Matching for Recognizing Natural Scene Categories

In line with the boom of 3D movies and cutting edge technologies, range cameras are increasingly common. Among others, time-of-flight (TOF) cameras give it the ability to capture three-dimensional images that reveal object's distances. A shortcoming of these sensors however, lies in that the majority does not provide color information (not even gray). Therefore they are useless in computer vision applications for which color is crucial. The PMD [vision] ® CamCube 3.0 is one example of an expensive colorless TOF camera. In this work, we attempt the addition of color to this camera by means of inexpensive resources. A regular webcam is stuck on top of the CamCube and its color images are registered into the TOF distance maps. To get this done, we developed an algorithm to enable real-time registration based solely on depth. Thus, this algorithm requires neither intrinsic parameters nor mono-calibration of none of the cameras. We finally show a tracking application in which a stumble is foretold if an object approaches following a threatening trajectory.

/pdf/real-time-image-registration-of-rgb-webcams-and-colorless-3d-4q1gpyrsbr.pdf

Real-time image registration of RGB webcams and colorless 3d time-of-flight cameras

Research in Brain-computer interaction, Multimodal Interaction Group, Computer Vision and Multimedia Laboratory, University of Geneva

In functional magnetic resonance imaging (fMRI), modeling of the complex link between neuronal activity and its hemodynamic expression via the neurovascular coupling usually requires the use of elaborated models. To avoid linear assumptions and a priori modeling of this expected hemodynamic signal, Bayesian approaches have recently improved their accuracy in estimating and detecting brain activation. Recent studies, using Markov random field (MRF) to represent activated brain voxels and likelihoods to find the maximum a posteriori (MAP) estimation of model parameters, provided superior efficiency in comparison with the context-free and the statistical parametric mapping (SPM) approaches. We propose another approach for detecting brain activity by introducing non-Gaussian and non-stationary Gaussian image models that exploit local subband image statistics in the non-decimated wavelet domain. These statistical models, being very simple and tractable, have demonstrated state-of-the-art performance in a number of applications including lossy image compression, denoising and digital watermarking. Such an approach yields close form analytical solutions with low computational complexity that makes it very attractive in neuropsychological research allowing to obtain theoretical performance limits. These models in the scope of the Bayesian estimation framework are applied on synthetic fMRI data corrupted by an additive white Gaussian noise (AWGN) with varying variance and on real fMRI data obtained from a motor preparation task. Comparison of the results with those obtained using SPM maps demonstrates the high efficiency of the proposed method.

Non-stationary Gaussian spatio-temporal modeling of fMRI data

Nowadays, there are many different types of mobility aids for elderly people. Nevertheless, these devices may lead to accidents, depending on the terrain where they are being used. In this paper, we present a robust ground change detector that will warn the user of potentially risky situations. Specifically, we propose a robust classification algorithm to detect ground changes based on colour histograms and texture descriptors. In our design, we compare the current frame and the average of the k previous frames using different colour systems and Local Edge Patterns. To assess the performance of our algorithm, we evaluated different Artificial Neural Networks architectures. The best results were obtained by representing in the input neurons measures related to Histogram Intersections, Kolmogorov-Smirnov distance, Cumulative Integrals and Earth mover's distance. Under real environmental conditions our results indicated that our proposed detector can accurately distinguish the grounds changes in real-time.

Thierry Pun

Papers

Real-time image registration of RGB webcams and colorless 3d time-of-flight cameras

Research in Brain-computer interaction, Multimodal Interaction Group, Computer Vision and Multimedia Laboratory, University of Geneva

Non-stationary Gaussian spatio-temporal modeling of fMRI data

A robust, real-time ground change detector for a “smart” walker

MRML: A communication protocol for content-based multimedia retrieval