METRIC LEARNING FOR LARGE IMBALANCED FACE DATASETS

Abstract

Computer vision is a very trending field nowadays. The amount of digital data i.e. image, audio, video etc. is increasing day-by-day at a faster rate. So, various algorithms are being developed around such digital payloads to extract the maximum potential of computer systems. In this project I’ll be working on image analysis, how these digital images can be played with on computer systems, how to remodel them based upon certain distinct characteristic features and its classification using various classifiers and similarity metrics like SVM, cosine similarity. I’ll also be using Metric learning after running inbuilt functionalities, transforming the image parameters and again doing performance analysis of classification which leads out to be better than the initial results. I’ll also be using Deep neural networks for feature extraction only and then apply my procedure for classification. I will be trying to devise a new algorithm to help improve the performance metrics for large imbalance datasets.