In the modern world, billions of sensors are continuously generating “signals” — streams of raw numerical data. Theoreticians classify signals by dimension:
1D Signals: Sound, temperature, pressure, seismic, prices, …
2D Signals: Images, spectrograms, topography …
3D Signals: Video, MRI, CT scans, …
4D Signals: 4D CT, ultrasound, …
The raw data in these signals needs to be converted to intelligent information. This is a difficult task, involving hundreds of diverse mathematical techniques. The diversity and complexity of the task at hand is compounded by the diversity of signal characteristics, and the needs of the application that needs the information to be extracted.
The ubiquity of data sources and their
near-universal importance in informing decisions and
creating new technology implies that
signal processing
is a part of almost any advanced engineering
project
We have worked on diverse signal processing projects, using a very wide spectrum of signal processing techniques:
• Fourier, wavelet, Laplace transforms
• Optimal, statistical and adaptive signal processing
• Control systems theory
• System identification theory
• Non-linear signal processing
• Topological signal processing
• A.I. techniques such as convolutional deep networks
• Image flow, disparity and other image processing techniques
• Compression of signals, images, video
• Graph algorithms such as max-flow, applied to nD signal processing
EXAMPLE PROJECTS
Oil & Gas Production
Thousands of sensors on an oil field are continuously generating data. This data may have missing values, wrong values, noise, and may be sampled at different times and rates. Generating high quality data is important before this data can be used to understand more about the physics of the underlying reservoir and production network.
Using data science for oil & gas production
Econometrics
Econometrics is the application of signal processing and statistical methods to economic data. Such techniques are of theoretical as well as practical importance. Practically signal processing and data science techniques can be used to build quantitative market models. This helps answer questions like pricing and inventory strategies.
Data science for managerial economics
Human Guided Machine Vision
Our HGMV system combines real-time image processing and machine vision algorithms with an advanced GUI for instantaneous human feedback. It can deliver real-time, precise, and contextual results in many areas including in metrology, medicine, film restoration, special effects, archeology and more.
Read more about the algorithms and applications
Denoising for Hearing Aids
Hearing aids help the aged and the hearing impaired lead close-to-normal lives. To achieve the best results, sound must be presented to the wearer with maximum clarity and minimum noise or interference. Furthermore, this has to happen in real time.
Read about our research and patent on hearing aids
Video Extensometry
To understand mechanical behavior of objects, it is important to measure stresses and strains. One way of measuring strains is by using video processing techniques called image flow analysis. Accurate strains can be calculated all over the object being deformed.
Using this in material testing
Video Codecs
We have worked on many video encoders and decoders (codecs), including standard-based codecs such as MPEG and H.264 and bespoke video codecs for specialized applications. In some cases, we have used the processing power of multi-core processors to increase codec performance while saving power consumed.