| 1995 |
Image Filtering
We conducted a research program on various aspects of image
filtering (standard, and some custom schemes). User application for tests
and demonstrations was implemented. |
| 1996 |
Efficient approaches to rendering of triangulated
surfaces
Rendering of triangulated 3D scenes is a complex and general
task. However, several problems may be eliminated if we consider
triangulated connected surfaces. Considering only connected surfaces (for
example, surface of a globe) we can efficiently implement rendering
procedure. In many cases, texture filtering may be done on-fly incrementally,
and Z-buffer is not required at all. We made an extensive research on the
above mentioned topics. |
| 1996 |
Efficient progressive compression of very large
images
It is always needed to view just a part of an image if the
image is very large (such as a map or a plan). We have developed
hierarchical approach for compression of large images, that allows to
extract image parts in required details level. Our approach allows to
achieve fast and efficient zoom/pan features on very large images. |
| 1997-1999 |
Efficient skeletonization technique
Our team invented very efficient solution for conversion of
raster data into vector representation. We used special relief methods
based on non-Euclidian metric, and scan-line approach. As a result, we
have developed an algorithm which is; (1) very fast (processing time
depends linearly on horizontal connectivity, but not on image dimensions);
(2) reversible, that means that the original image can be constructed from
obtained skeletons. We also developed several methods of post-processing,
such as simplification and defects filtering.
Based on our unique method, we developed plug-ins for AutoCAD Map
(AutoDesk, Corp.) and GeoMedia (Intergraph, Corp.) |
| 1998-1999 |
Hierarchical Techniques of Image Compression
We have developed a new approach to image compression which
was based on hierarchical structures, in particular, on binary trees. This
approach provides progressive image storage/transmission. Potentially
designed for lossless compression, proposed scheme appeared to be superior
over other standard techniques (such as PCX, GIF, PING, ...) in terms of compression
ratio. The other important characteristic is extremely fast and simple
compression and decompression, which both allow parallel computations
provided by SIMD instructions.
This project was conducted under Research Agreement with
Intel
Technologies, Corp. |
| 1999 |
Texture Compression Technique for 3D Pipeline
We have developed a new technique of texture compression that is
superior over other existing techniques (S3TC, DXT, FXT1) in quality and
compression ratio. This technique was extensively tested in pipeline
simulator.
This project was conducted under Research Agreement with
Intel
Technologies, Corp. |
| 2000 |
New approach to 3D data representation and
rendering
The project included research and development of a new technique, which
turned out to be superior over standard approaches
for 3D scenes representation, such as polygonal approaches. This technique
is based on native representation of real 3D objects, but provides very
efficient and uniform rendering, which even does not require specially
built accelerators in many cases.
This project was conducted under Research Agreement with
Intel
Technologies, Corp. |
| 2001 |
Analysis and Improvement of Range
Images
The project includes research and development of automated system for 3D
model acquisition. The initial data are range images taken by 3D scanning
machinery. Various shape representations, including the one developed in
2000, are considered as the resulting structure. Techniques of data
improvement by statistical estimation utilizing Bayesian approach are
studied. The second phase of the project is dedicated to reconstruction of
range data from multiple plain images.
This project is conducted under Research Agreement with
Intel
Technologies, Corp. |
| 2001-2003 |
Head Model Calibration and Reconstruction
The project includes research and development of automated framework that
would allow for calibration of a generic head model from several images or
video.
This process implies geometrical deformations of the respective meshes
(such as skin, eyes, etc.) and generation of an appropriate texture. The
complete calibration pipeline is designed to meet requirements of MPEG-4
standard for Calibration and Predictable Facial Animation Object profiles.
This project is conducted under research agreement and in cooperation with
Intel Nizhny Novgorod Lab (Laboratory of Intel Corp.). |
| 2003 |
Statistical Classification of Raster Images
The aim of this research is application of machine learning and
probabilistic networks for the purpose of raster image classification and
further processing, e.g. corresponding data compression.
This project is conducted under research agreement and in cooperation with
Intel Russian Research Center (Intel Corp.). |
| 2003 |
Making video from still images
The project aims at developing a utility for making video stream from
still images of high resolution. This utility allows for making zoom-in/out,
pan and other navigational operations in the image.
This project is conducted for Laboratory of Photogeology of the
Department of Geology of Moscow State University. |
| 2004 |
Temporal Video Segmentation for
Video Mining
The project aims at providing efficient in terms of quality and
computational costs mechanism for temporal segmentation of a video stream
into separate shots, i.e. an unbroken sequence of frames related to one
scene and taken by one camera. Detection of single camera operations (i.e.
zoom, pan, tilt, etc.) is considered as an important step for further
temporal analysis of the video. Resulting segmentation and related
characteristics are one of the basic elements of video mining engines.
This project is conducted under research agreement and in cooperation with
Intel Russian Research Center (Intel Corp.). |
