Neural Receptive Fields Research Articles

Entropy reduction and decorrelation in visual coding by oriented neural receptive fields.

Image coding is optimally efficient if the coefficients of the representation are nonredundant, in the Shannon sense that their mutual information is zero. In that case, the code coefficients are uncorrelated and they form a statistically independent ensemble, so that the conditional probability of one coefficient's value, given any other coefficient's value, is the same as its unconditional probability: P(x/y) = P(x). In order for each coefficient to capture a unique property of the image that cannot be captured by any other coefficient, the expansion functions employed in the code must be linearly independent. In order for the code coefficients to have zero mutual information, the code primitives must be orthogonal so that their projections onto each other are always zero. In biological visual systems, although it is clear that some forms of efficiency (such as speed) are desirable, it is not obvious whether coding efficiency as measured by mutual information among the neurons is a factor which explains any of their properties. The center/surround receptive field profiles of neurons in the retina and geniculate are far from an orthogonal set, but a given neuron can still be regarded as a decorrelator of the incoming signal in the sense that it responds primarily to changes in the image (changes in space, time, chrominance, etc.) At the level of the brain's visual cortex, the introduction of the new variable of orientation selectivity can be regarded not only as a means for providing orientation labels for image structure, but also more basically as an effective decorrelator of the neural representation.(ABSTRACT TRUNCATED AT 250 WORDS)

Understanding the efficiency of human perception.

We compare the high perceptual efficiency exhibited by humans at a number of tasks in visual perception with that predicted by models of visual information processing, in particular the broad class of "feature detector" models, which we represent as an Ising spin system responding to a quasirandom magnetic field. The observed excitatory-center, inhibitory-surround organization of the neural receptive fields is seen to be desirable.

Pattern and motion vision without Laplacian zero crossings.

The zero crossings of band-limited signals are known to be rich in information. Recent models of information processing in biological visual systems have proposed that image structure is represented initially by such zero crossings, after the image has been bandpass filtered at multiple scales by neural receptive fields whose two-dimensional profiles resemble the Laplacian of a Gaussian. Because the resulting zero crossings generally correspond to physically meaningful structures such as edges and occlusion boundaries, and indeed under some conditions can exhaustively specify the original image, such representations have proven useful and efficient in a variety of machine vision problems. However, some simple information-processing operations that are apparent in human pattern and motion vision can be shown to be impossible in such representations, because the zero crossings in the bandpassed signals do not capture the necessary information (at any scale of analysis), because the information that they provide is misleading, or because there are no such crossings in the signals after delta 2 G sigma filtering at any scale. Examples are provided of simple visual signal-processing tasks (texture discrimination, motion perception, pattern detection) that human beings can perform effortlessly but that cannot be performed in the proposed multiscale delta 2 G sigma zero-crossings schemes. These perceptual capabilities offer evidence against this model of early image representation in human vision.

Six formal properties of two-dimensional anisotropie visual filters: Structural principles and frequency/orientation selectivity

Six formal properties of anisotropic linear two-dimensional filters are noted which are relevant for modeling the mechanisms of spatial visual information extraction. These properties concern the relationship between the organizational principle of a two-dimensional anisotropic spatial filter or neural receptive field (such as elongation, or concatenation of subunits, or differential operators mediated by lateral inhibition in neural laminae) and the resulting general consequences for spatial frequency and orientation selectivity. These properties are demonstrated without assuming particular two-dimensional filter functional forms; rather, they are shown as general principles associated with certain broad categories of two-dimensional filters. Such an analysis enhance theoretical understanding of the two-dimensional receptive field organization of neurons in the visual cortex and permits the rejection of some candidate organizational principles on two-dimensional spectral grounds.