Surveillance Robots based on Pose Invariant Face Recognition Using SSIM and Spectral Clustering

Abstract

In the current techno-savvy world, cybersecurity is a prime concern. Biometrics are being extensively used for authentication and authorization. Face recognition(FR) is a class of biometrics which has proved to be one of the most effective methods for identification and verification, which works even when the subject is unaware of being scanned. This paper gives insight into the use of Image Processing in Robotic Applications. It discusses the use of computer vision by surveillance robots. Surveillance applications prefer unsupervised learning over supervised learning as unsupervised learning doesn’t require true labels. The burgeoning demand for unsupervised learning in surveillance applications proffered the nub of this project. The paper coalesces the well-known similarity algorithm SSIM with Spectral Clustering to produce prodigious results. SSIM surpasses other techniques like MSE by extracting structural features from images. This leads to a significant improvement in performance because humans also extract structural information from visuals. SSIM eliminates the effect of illumination and then uses the attributes that depict the structure of objects to gain the desired structural information. The performance of the proposed model was compared with other similarity measures on the ORL(Olivetti Research Laboratory Cambridge), Caltech and Faces96 datasets. An accuracy of 89.5% was achieved on the ORL and 88.3% on the Caltech and 86.7% on the Faces96 dataset.