PROCESSING AND COMPRESSION
MULTIMEDIA INFORMATION - IMAGES, VIDEO AND AUDIO
Lossless compression
Applying knowledge about the structure of images, video and sound, about the formats of their injury in conjunction with coding algorithms, it is possible to achieve compression 3-6 times or more without losing a single bit of information.
Lossy compression
Multimedia information has the so-called perceptual redundancy, which allows you to delete part of the information so that it will be invisible to a person. In this case, it is necessary to process the data in such a way as to distinguish material and non-material information.
Methods and Technologies
We apply a wide range of technologies, from wavelet analysis to variational auto encoders. At the same time, the issues of metrics for assessing compression quality and multimedia quality are critically important, which also depend on specific scenarios. We use both standard types of SSIM and QoE, based on human perception.
PROJECTS
The Laboratory is developing a modern video codec based on multi-channel wavelet decomposition:
- Building a multi-band wavelet filter bank;
- The development of visual compression techniques;
- Combination Transformation Research;
- Development of entropy coding methods.
One of the areas of work is the use of neural networks to eliminate the occurrence and correction of distortions.
- Development of new methods for post-processing video and still images;
- The use of convolutional neural networks for the tasks of improving and lowering the quality;
- Reducing the impact of compression artifacts;
- Reduced Bitrate for Limited Bandwidth Channels.
Based on the competencies gained during the development of the video codec, post-processing of video and quality metrics that meet human perception, work has begun on the creation of methods and algorithms for compressing and decompressing video and images in ultra high quality.
- Deep knowledge in classic video processing;
- Codec knowledge;
- Bayesian Neural Networks;
- Post processing;
- Correct metrics;
- Interaction with industry experts.
Research, development and modernization of the open source speech codec “Codec 2”, including adaptation to Russian speech and tuning to a low-speed connection - reducing bitrate while maintaining speech intelligibility.
- Preservation of complex sounds such as hissing or deaf consonants;
- Modify effective noiseless shutter without reducing clarity;
- Improving data encoding and compression algorithms;
- Adding intermediate velocity values to provide a more accurate channel offset.