ISSN: 2320-5407 Int. J. Adv. Res. 4(9), 522-529 522 Journal Homepage: - www.journalijar.com Article DOI: Article DOI: 10.21474/IJAR01/1513 DOI URL: http://dx.doi.org/10.21474/IJAR01/1513 RESEARCH ARTICLE PRIVACY PROTECTED FACE VERIFICATION SYSTEM USING SPARSE CLASSIFIER. Thasni A N and Deepthi V R Department of Computer Science and Engineering, RIT, Kottayam, India. …………………………………………………………………………………………………….... Manuscript Info Abstract ……………………. ……………………………………………………………… Manuscript History Received: 12 July 2016 Final Accepted: 19 August 2016 Published: September 2016 Key words:- face scrambling; SIFT; LBP; T-test; Sparse Representation classifier. The demand of video surveillance has been increasing day by day; as a result the privacy protection has become a responsibility for the public as well as for legal authorities. When videos are transmitted or distributed across various public networks, the human faces should not be exposed. To deal with this problem, facial image scrambling technique appeared as a solution for privacy related applications. This paper proposes a facial verification system in the scrambled domain using sparse classifier. In the proposed method, the facial features are extracted from the scrambled faces using SIFT and LBP feature extraction methods and a T-test based feature selection method is used to select important features for classification. Then sparse representation based classifier is used for classifying the facial images. The experiments show that the proposed face verification system can meet the challenging tests in the scrambled domain. Copy Right, IJAR, 2016,. All rights reserved. …………………………………………………………………………………………………….... Introduction:- Today visual surveillance has become an extensively used technology in various applications. Exposing of face images [1, 2] in the surveillance videos should be avoided when it is transmitting and distributing over different networks. One of the solutions to this problem is the Scrambling technique [3] where the privacy of the subjects can be respected. Image scrambling give more advantages than other methods, it provides lower computational cost and time than encryption method. Scrambling is a popular method in visual surveillance. It does not really hide anything from surveillance video; it can only avoid exposure of human faces from surveillance videos. This paper uses scrambling of faces using Arnold transform [4, 5]. Using inverse Arnold transform scrambled faces can be recovered easily with different parameters. There are many scrambling techniques are available. For example, scrambling can be performed by using cartooning techniques [6]. After this method, the faces become extremely hard and it lost the facial information. As a result the face recognition/verification is not performed successfully in this case. So this method is not a good choice for hiding human faces from surveillance video. The Arnold transform [4, 5] differ from all other methods, it is a recoverable type scrambling method and it only change the pixel position. Hence this work uses Arnold transform based scrambling method. The automatic face verification/recognition is usually a challenging task. Due to this reason face recognition has become a significant research topic in indexing of images [4], human-computer interaction [7~9], forensic biometrics [10] and medical applications [11]. In visual surveillance system, the captured videos are transmitted on Corresponding Author:- Thasni A N. Address:- Department of Computer Science and Engineering, RIT, Kottayam, India.