Research Areas
A unifying theme in my research is the effort to understand and computationally formalize biological information processing
systems (like the Visual and Auditory systems), and then use this gained knowledge to design and improve artificial
computerized systems sharing the same goals.
Computational principles behind Cortical Architecture: We provide evidence that the visual system has an initial stage,
when visual information is estimated on a grid whose spatial resolution is exactly that of the classical Hubel& Wiesel hypercolumns.
We have shown that cortical ocular dominance columns might play a crucial role in processing binocular data, and proposed a
Cepstral algorithms that can efficiently utilize this cortical “data structure”(PAMI89). We then focused on the role of the ocular
columns in early vision, and found out that the visual system seems to initially assigns a single value to each visual dimension
(i.e. depth or motion vector) in image patches whose size is about 10 minutes of arc, which coincides with the foveal size of the
visual hypercolumns, and also found this spatial patch size to scale according to the cortical magnification factor
(Perception95,SpatialVision98,BiolCyb99).
Computational principles common to the Visual and Auditory systems: Looking for common processing principles behind
various sensory modalities, we have suggested that the Auditory system is using an “Auditory Edge Detection” units, very similar
to the ones used by the visual system (J.NeuroPhys01). We then provided evidence that such units might be spatially structured
in a way that is reminiscent of the classical visual columns.(J.NeuroPhys03) , see an editorial focus on our work.
Foveated vision and Detection of Regions of Interest: Following the foveated, log-polar based, visual information principle
employed by the human system, we explored various methodologies for automatic detection of regions of interest in images.
We have suggested two main approaches – an edge based generalized symmetry operator (ICCV90,IJCV95,CVIU98), and a
direct, image intensity based method that is looking for traces of 3D convex objects in the 2D image(BMCV00, CVIU01, ,CVPR01).
In this regard, we have also suggested to assess 3D shape of objects by using the perspective projection rather than by the commonly
used orthographic one (ICCV03 ,CVPR04 )
Functional MRI: We are starting to employ fMRI as a tool that provides a glimpse into the “black-box”. Designing a non-orthodox
experiment based on a free game played between the experimenter and the subjects, we were able to tell, just by analyzing patterns of
activity in the human amygdala, whether subjects were cheating during the game(Neuron02). Looking further into the realms of "intention"
and
"Free will", we explore the brain networks involved in volition and
decision (JNS)