Machine Analytics & its Application
(May 14th, 2018 – June 30th, 2018)
Organized as a TEQIP Summer Training Program (STP) 2018, the STP covered topics in Machine Learning and was designed to provide the participants with an experience to implement the ML techniques on different projects.
Improving the images captured in dark and a new learning algorithm
Capturing the images in dark has always been difficult due to less SNR. An image processing pipeline is the set of components commonly used between an image source and an image renderer. Lot of development is done in the technology of camera sensors, but the image processing pipeline is still left behind. With this project we tried to develop a new pipeline to capture the images in the dark. The work is still in progress and we are using convolution neural network in our pipeline. A training algorithm is the process which carries out the learning procedure in neural networks. There are many different learning algorithms with different characteristics and performance. We have attempted to introduce a new algorithm and have named it ad Gain Root Mean Square(GRMS).
Emotion Detection and Characterization using Facial Features
In this project, a model is built that aims to detect an emotion from any given input image and deploys the emotion characteristics from images of human faces available in Cohn-Kanade dataset using machine learning algorithms. The dataset was initially segregated and organized according to their labels. The eyes and mouth were then detected and extracted using the Viola-Jones Algorithm and passed through the Gabor filter. The resulting data was further converted to pixel arrays, split into training data (80%) and testing data (20%), and eventually passed through a Support Vector Machine along with their respective labels for classification. The testing data returned a classification accuracy of 81%, at its latest.
Handwritten Character Recognition
Handwritten Character Recognition is one of the most active research areas in the field of Image Processing and Pattern Recognition. It has numerous applications which include reading aid for blind, language translation, bank cheques and conversion of any handwritten document into structural text form. The aim of theproject is to recognise handwritten English uppercase letters using Neural Networks and Feature Extraction. The images of the dataset were subjected to preprocessing which involved steps like skew correction, binarization and resizing the image. The Histogram of Oriented Gradients feature extraction method has been used to extract the relevant features from the image. k- Nearest Neighbor is used as a classifier for the features. The dataset taken was from NIST handwritten dataset 2016. For 20,800 test cases, our program recognised 87% characters accurately.
Access Control using Facial Recognition
A system which can detect and recognize face through a camera/image/video. We have implemented it using several methods. First using Haar Cascades, second by using tensor flow, third by creating our own Haar Cascade and forth by Backpropagation Network. All the programs are implemented using Python 3. We are currently working on facial recognition using (BPN) and trying to increase its accuracy.
The LNM Institute of Information Technology,