Directional selectivity to local visual stimuli appears in various levels of the visual pathway, being in the retina very conspicuous. Neurophysiology suggests that directionality (as well as other local and quasi-global filtering properties) are based in the space–time interactions of processes with different “memory” (latency). We draw inspiration from the corresponding underlying biological mechanisms to propose two general schemes for directionality computation in nets, compatible with other space–time filtering properties. First, a connectivistic mechanism based on bipolar–amacrine–ganglion cell interaction is proposed, by formalizing the classical proposals of early vision neurophysiologists. Second, inspired initially in the more recently described intrinsic directionality of amacrines, novel schemes are proposed where directionality appear as the computing consequence of adding memory to spatial filtering structures. The mathematical formulations are achieved by means of Newton Filters and Hermite Functionals.

Bioinspired computing nets for directionality in vision

1)

We have studied in depth the Theory of Receptive Field Functinal Transformation, increasing their scope to situations of variable dimension partitions.




2)

We have developped a theorem which allows the Systematic generation of Complete Transform in the above case.

Cast Methods for Generation of Non-Orthogonal Complete Transforms

The interplay between the design of layered computing systems and the study of real visual structures and functions is a very fruitful area in which the characteristics of the natural systems inspire properties and design criteria of the artificial counterparts. Visual saliency and attention mechanism are two of the most explored and interesting features of the visual processing in higher vertebrates. In general, visual saliency refers at the propensity of some visual features to attract attention, while attention by itself requires a voluntary effort to locate places of interest in the visual scene. In this paper we present a study of a mechanism of attention that has variable selection criteria based in highlighting some feature or object in the image while blurring the rest of it. The fact that all the information is preserved in the process, allow us to be able to select a new location of interest in a efficient way. This mechanism is based in a class of transform which combines partitions of the data field and functionals acting on them and is suitable to perform a selection process of the attended location with different levels of resolution.

A Bio-inspired Proposal for Focus Attention While Preserving Information

1)

We have studied in depth the Theory of Receptive Field Functional Transformation, increasing their scope to situations of variable dimension partitions.




2)

We have developed a theorem which allows the systematic generation of Complete Transform in the above case.

Towards a CAST Method for a Systematic Generation of Non-Orthogonal Complete Transformations

A fundamental concept in Artifitial Vision and Image Processing is that of Complete Description of a data field. From the analitycal point of view completness requires the conservation in the number of degrees of freedom of the visual environment. This constancy has allowed us to establish a special conservation principe, which is determinated by the receptive fields and the functional performed on them.

The Role of Partitions and Functionals in Descriptor Computation for Data Receptive Fields

Presynaptic Inhibition (PI) basically consists of the strong suppression of a neuron’s response before the stimulus reaches the synaptic terminals mediated by a second, inhibitory, neuron It has a long lasting effect, greatly potentiated by the action of anaesthetics, that has been observed in motorneurons and in several other places of nervous systems, mainly in sensory processing In this paper we will focus on several different ways of modelling the effect of Presynaptic Inhibition(PI) in the visual pathway as well as the different artificial counterparts derived from such modelling, mainly in two directions: the possibility of computing invariant representations against general changes in illumination of the input image impinging the retina (which is equivalent to a low-level non linear information processing filter) and the role of PI as selector of sets of stimulae that have to be derived to higher brain areas, which, in turn, is equivalent to a “higher-level filter” of information, in the sense of “filtering” the possible semantic content of the information that is allowed to reach later stages of processing

Juan Carlos Quevedo-Losada

Papers

The Role of Partitions and Functionals in Descriptor Computation for Data Receptive Fields

Cast Methods for Generation of Non-Orthogonal Complete Transforms

Systems Approach to Attention Mechanisms in the Virtual Pathway

Towards a CAST Method for a Systematic Generation of Non-Orthogonal Complete Transformations