Yehezkel Yeshurun

Context-free attentional operators: the generalized symmetry transform

Active vision systems, and especially foveated vision systems, depend on efficient attentional mechanisms. We propose that machine visual attention should consist of both high-level, context-dependent components, and low-level, context free components. As a basis for the context-free component, we present an attention operator based on the intuitive notion of symmetry, which generalized many of the existing methods of detecting regions of interest. It is a low-level operator that can be applied successfully without a priori knowledge of the world. The resultingsymmetry edge map can be applied in various low, intermediate-and high- level tasks, such as extraction of interest points, grouping, and object recognition. In particular, we have implemented an algorithm that locates interest points in real time, and can be incorporated in active and purposive vision systems. The results agree with some psychophysical findings concerning symmetry as well as evidence concerning selection of fixation points. We demonstrate the performance of the transform on natural, cluttered images.

Image warping by radial basis functions: applications to facial expressions

Abstract The human face is an elastic object. A natural paradigm for representing facial expressions is to form a complete 3D model of facial muscles and tissues. However, determining the actual parameter values for synthesizing and animating facial expressions is tedious; evaluating these parameters for facial expression analysis out of gray-level images is ahead of the state of the art in computer vision. Using only 2D face images and a small number of anchor points, we show that the method of radial basis functions provides a powerful mechanism for processing facial expressions. Although constructed specifically for facial expressions, our method is applicable to other elastic objects as well.

Widespread functional connectivity and fMRI fluctuations in human visual cortex in the absence of visual stimulation.

To what extent does the visual systems activity fluctuate when no sensory stimulation is present? Here, we studied this issue by examining spontaneous fluctuations in BOLD signal in the human visual system, while subjects were placed in complete darkness. Our results reveal widespread slow fluctuations during such rest periods. In contrast to stimulus-driven activity, during darkness, functionally distinct object areas were fluctuating in unison. These fMRI fluctuations became rapidly spatially de-correlated (39% drop in correlation level, P < 0.008) during visual stimulation. Functional connectivity analysis revealed that the slow spontaneous fluctuations during rest had consistent and specific neuro-anatomical distribution which argued against purely hemodynamic noise sources. Control experiments ruled out eye closure, low luminance and mental imagery as the underlying sources of the spontaneous fluctuations. These results demonstrate that, when no stimulus is present, sensory systems manifest a robust level of slow organized fluctuation patterns.

Superresolution in MRI: Application to human white matter fiber tract visualization by diffusion tensor imaging

A superresolution algorithm was applied to spatially shifted, single‐shot, diffusion‐weighted brain images to generate a new image with increased spatial resolution. Detailed two‐dimensional white matter fiber tract maps of the human brain resulting from application of the technique are shown. The method provides a new means for improving the resolution in cases where k‐space segmentation is difficult to implement. Diffusion‐weighted imaging and diffusion tensor imaging in vivo stand to benefit in particular because the necessity of obtaining high‐resolution scans is matched by the difficulty in obtaining them. Magn Reson Med 45:29–35, 2001.

Robust detection of facial features by generalized symmetry

Locating facial features is crucial for various face recognition schemes. The authors suggest a robust facial feature detector based on a generalized symmetry interest operator. No special tuning is required if the face occupies 15-60% of the image. The operator was tested on a large face data base with a success rate of over 95%.<<ETX>>

Cepstral filtering on a columnar image architecture: a fast algorithm for binocular stereo segmentation

Many primate visual cortex architectures have a prominent feature responsible for the mixing of left and right eye visual data: ocular dominance columns represent thin (about 5-10 minutes of arc) strips of alternating left and right eye input to the brain. It is shown that such an architecture, when operated upon with a cepstral filter, provides a strong cue for binocular stereopsis. Specifically, the vector of binocular disparity may be easily identified in the output of the (columnar based) cepstral filter. This algorithm is illustrated with application to a random dot stereogram and to natural images. The authors suggest that this provides a fast algorithm for stereo segmentation, in a machine vision context. In a biological context, it may provide a computational rationale for the existence of columnar systems with regard to both ocular mixing and other visual modalities that have a columnar architecture. >

Detection of interest points using symmetry

An operator based on the intuitive motion of symmetry, which effectively locates interest points in real time and can be incorporated also in active visual systems, is introduced. The results of its operation agree with some psychophysical evidence concerning symmetry as well as evidence concerning fixation points. The operator can be applied successfully without prior knowledge of the world. Combining the operator with some preconceptions about the image is a powerful tool for feature detection in intricate natural scenes. The localization of faces and facial features in real time is demonstrated on detailed and noisy pictures.<<ETX>>

Shape-from-Shading Under Perspective Projection

Shape-from-Shading (SfS) is a fundamental problem in Computer Vision. A very common assumption in this field is that image projection is orthographic. This paper re-examines the basis of SfS, the image irradiance equation, under a perspective projection assumption. The resultant equation does not depend on the depth function directly, but rather, on its natural logarithm. As such, it is invariant to scale changes of the depth function. A reconstruction method based on the perspective formula is then suggested; it is a modification of the Fast Marching method of Kimmel and Sethian. Following that, a comparison of the orthographic Fast Marching, perspective Fast Marching and the perspective algorithm of Prados and Faugeras on synthetic images is presented. The two perspective methods show better reconstruction results than the orthographic. The algorithm of Prados and Faugeras equates with the perspective Fast Marching. Following that, a comparison of the orthographic and perspective versions of the Fast Marching method on endoscopic images is introduced. The perspective algorithm outperformed the orthographic one. These findings suggest that the more realistic set of assumptions of perspective SfS improves reconstruction significantly with respect to orthographic SfS. The findings also provide evidence that perspective SfS can be used for real-life applications in fields such as endoscopy.

Preprocessing of Face Images

The reliability, speed, and complexity of virtually any face recognition system are substantially improved if the location and the scale of the faces are known. We propose a method for automatic and robust detection of the eyes and mouth using the context freegeneralized symmetry transformand knowledge of faces. The features are extracted from the image of the intensities gradients and are then used to normalize the face images. We show that a normalization procedure based on affine transformations whose anchor points are the locations of the eyes and mouth substantially increases the effectiveness of general purpose classification techniques in face recognition.Other normalization procedures for avoiding the effect of background and varying light conditions are proved to be instrumental as well.

Face recognition using a hybrid supervised/unsupervised neural network

Face recognition schemes that are applied directly to gray level pixel images are presented. Two methods for reducing the overfitting-a common problem in high dimensional classification schemes-are presented and the superiority of their combination is demonstrated. The classification scheme is preceded by preprocessing devoted to reducing the viewpoint and scale variability in the data.