Head of Laboratory – Dr. Modest Vaintsvaig

Tel.: (095) 209-42-25; E-mail: wainzwei@iitp.ru

The leading researchers of the laboratory include:

In the course of the development of Spectral Stimuli Creation theory we build an algorithm for generating projectively invariant representation of plane compositions of contrast image elements. Being applied to the task of reconstruction of correspondence for boundaries obtained by single-image color segmentizer, the algorithm privides more reliable (compared to static one) recognition of occlusion countours and light spots. Also an algorithm is developed for estimation of the shape of moving object from two temporally sequential projections of their base points sets.

We developed some implementations of algorithm for establishing structural correspondence (association) between representations of dynamic scenes, where every dynamic object identifier is bound with a set of pieces of its 2.5D images, each being a projecton of the object onto fixed-size raster screen with its own resolution and other specific features. In each set, the pieces are supplied with pairwise transformations according to point and time of the observation. The general algorithmical framework for obtaining such representations from source images is developed and partly implemented.

The new version of algorithm for establishing point-wise correspondence between gray-scale images is implemented as a computer program. It allows for higher values of disparity derivatives than earlier versions, and is capable to deal with occlusions.

We developed algorithms that distinguish different (projectively non-euqivalent) convex smooth 2D-bodies whose central projections are given. This algorithm uses projective invariants and detects symmetry (axis symmetry, point symmetry and potational symmetry). It is robust with respect to space quantization errors.

We investigated optimal adaptive methods for solving linear operator equations with random error in the right-hand side. Using singular decomposition, we reduce the problem to finding an estimate for (infinite-dimensional) parameter in random nonuniform noise We suggest a method to get such an estimate by processing parts (blocks) of parameter value using midified Stein-James rule. We have proved that if blocks are weakly geometrically increasing and is Stein's estimate is penalized in a certain way, this method is adaptive for a broad family of function classes.

We studied combinatorial counterparts for inequalities involving Kolmogorov complexity and Shannon entropy, especially linear inequalities with positive coefficients and only one term in the left-hand side. We proved that all (true) inequalities of that type are linera combinations of basic inequalities and are true for prefix complexity up to $O(1)$-term.

We studied the possible application of chaotic mappings in queueing theory. We suggest a method to choose an adequate model based on ``synthetic analysis"

We have completed successfulle the joint project with Samsung Institute of Advanced Technologies . A software for color and texture-segmentation of images is developed.

• Russian Foundation for Basic Research (No. 99-01-00407): “Analysis, Recognition and Structural Representation of Moving Visual Scenes”.
• Russian Foundation for Basic Research (No. 99-04-48791): “The parallel mechanisms of the representation and recognition of 3D objects based on dual mapping of elements in the linear models of the color and spatial perception for peripheral level of visual systems.