Image and Video Processing

Image/Video Processing was/is predominantly considered to be a set of advanced signal processing algorithms applied to images/videos to convert them to a state whereby inferences can be made from them for various purposes.

Computer Vision, on the other hand, usually refers to the science of extracting meaningful information from images/videos (usually processed through various image/video processing techniques) which solves a well-defined business case. Computer Vision is used to solve “Hard” problems – problems for which well-defined algorithms cannot provide precise mathematical solutions in an acceptable time period.

Machine Learning (Deep Learning), today, is capable of replacing some Image Processing requirements and can solve a larger set of Computer Vision problems. However, a careful selection of tools and algorithms are required to solve any problem within a time and at a cost which is relevant for the commercial problem on hand.

Whirldata works with customers to understand the time-cost-accuracy trade-off and various other factors that may impact the nature of the solution. Once this understanding is clear, various carefully chosen combinations of image processing and feature extraction methods are used to develop a ground-up solution.

Whirldata's team is capable of traditional image/video processing approaches such as Scale Invariant Feature Tranformation (SIFT), Harris Corner Detection/Transformations, Speed Up Robust Feature(SURF), Oriented Fast Rotated Brief (ORB) and several others. In addition, Whirldata has extensively used Convoluted Neural Networks and various deep convolutional encoder-decoder architecture based tools like SegNet to evaluate time-cost-accuracy trade-offs for different commercial scenarios. In addition, every developer is proficient with Python (with numpy) and C++ and between the team a significant amount of knowledge base is available for calculus, liner algebra, probability and statistics.