Face Recognition using Depth Information

Abstract

Depth information based face recognition deals with reorganisation of a person by using its depth. It is popular because of the advantages of the depth images over general colour or gray scale images. But it is also challenging because of the less availability of the databases and research papers in this field. Depth faces can be recognised using almost all two dimensional feature based methods like Local Binary Pattern (LBP) and others. Since we are more concerned about detecting the variation of the shape of images, modified versions of the LBP are used here. But these modifications come at the cost of adding some extra bits to the extracted features. Some of these modified versions of LBP are Depth Local Binary Pattern (DLBP), Depth Local Quantised Pattern (DLQP) etc. DLBP introduces a variable threshold along with LBP bit whereas DLQP perform quantification of the depth difference. The extracted features are normalised before applying to a classifier. The normalised feature descriptors have been classified using a Support Vector Machine (SVM) classifier. SVM classifier performs multi-class classification using one vs. all technique. It draws a decision boundary which is at maximum distance between the classes. All experiments have been done on HRRFaceD database. It is available online and has been acquired by Kinect 2 sensor. The results obtained from different have been shown in tabular form. The recognition rates are quite acceptable. Although, they doesn’t outperform the popular two dimensional techniques. But depth information can help recognise people in dark places.