Projects
Super-resolution of Images replacing convolutional neural networks
It has been seen that convolutional neural networks take a lot of time to implement and run very slow for on the go use. Thus, we are developing an alternative algorithm, using polygonal neural networks, which is supposed to perform faster.
Micro expression detection : Creating a framework for panel interviews
Using Voila-Jones Haar Classifier I detected major facial features that contribute to expressions. Then trained the Haar Classifier to the terminal points (eyes, nose etc.) and with the help of Canny Edge detector and Harris Corner Detection detected the external features (Face edge points,Ears etc.) Using 4 emotions: happy, sad, surprised, and neutral, trained the machine on several different examples.
Replacement for CCTV cameras
To replace the large amount of manpower and digital space required for recording different CCTV footage, by using OpenCV and python. I created a delta and a gray-scale image of the present frame and then thresholded the frame delta image to segment the foreground from the background.Then created an average image of the foreground and subtracted present image from it. As soon as movement was detected a camera would start recording and stop as the motion stops.
Projects on Kaggle.com
For the past 2 years I have been part of many individual as well as team projects based on unsupervised learning, fuzzy logic, regression and statistics on Kaggle.com and have been ranked in the top 10% twice. The projects were hosted by organisations like Aribnb, FICO, AXA, Wallmart, Homesite etc.