An Active Appearance Model based Face Recognition from Surveillance Video

Authors

  • T.Shreekumar, K.Karunakara

Abstract

For the past few years, the Face Recognition (FR) is becoming active research area in the field of Face Identification and verification. The Deep convolutional neural networks (CNNs) are extensively becoming popular in the field of FR in recent years but works with labelled datasets containing very large number of training samples. It is also difficult task to collect large number of Face images for training the model. Decision tree (DT) and-Nearest Neighbour (K-NN) performs better when the training set is small and computationally expensive when the training samples increases. In order to overcome the problem of low recognition and high computation complexity of Face Recognition (FR) space, this research paper proposes a Support Vector Machine (SVM) based FR to recognize the faces from video frames and still images. During the SVM training, the   parameters  are optimized with particle swarm optimization (PSO) technique , which enhances the FR rate.  In this method initially, the noise is eliminated from the probe image using Adaptive median Filter (AMF) and then the feature vector is generated using the combination of Active Appearance Model(AAM) and the shape model.  The recognition performance is analyzed on UPC Video Database, YouTube Face Database(YTF),ORL database and Yale B face sets . The main application of this research is to identify Faces from Poor quality surveillance video as well as still images.

 Keywords:Face Recognition, Support Vector Machine, Adaptive Median Filter, Particle Swam Optimization

Downloads

Published

2020-05-18

Issue

Section

Articles