International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395 -0056
Volume: 04 Issue: 03 | Mar -2017
p-ISSN: 2395-0072
www.irjet.net
Night time face recognition at large standoff Sonali Chhapre, Priti Jadhav, Ankita Sonawane, Paresh Korani 1Student,
Dept. of Computer Engineering, KKWIEER College, Maharashtra, India. Dept. of Computer Engineering, KKWIEER College, Maharashtra, India. 3Student, Dept. of Computer Engineering, KKWIEER College, Maharashtra, India. 4 Student, Dept. of Computer Engineering, KKWIEER College, Maharashtra, India. ---------------------------------------------------------------------***--------------------------------------------------------------------2Student,
Abstract - The face recognition in the night time situation
cross-spectral (NIR probe to VIS gallery) and cross-distance (60 m, 100 m and 150 m probe to 1 m gallery) face matching are desirable in practical applications.
becomes difficult, if the subject is at larger distance from the camera. Thus image quality degrades because of large standoff and low luminance at night time. To address this challenging issue of night time face recognition, an Augmented Heterogeneous Face Recognition (AHFR) approach is proposed. This approach is useful for cross-distance and crossspectral face matching. In this project, high quality face images are recovered from degraded probe images using an image restoration method based on Locally Linear Embedding (LLE). Further this restored image is matched with images in gallery or database using a heterogeneous face matcher.
The cross-spectral (NIR vs. VIS) and cross-distance (60 m, 100 m, and 150 m vs. 1 m standoff) face matching problem for the nighttime face recognition at large standoff are addressed here . Here the term heterogeneous in the context of face images refers to the images captured in both daytime and nighttime at different distance (e.g., 1 m, 60 m 100 m, and 150 m). For example refer fig 1 given below which depicts NIR images that are captured at night time at a large standoff (e.g., 150 m) still contain some discriminative facial details. A Long Distance Heterogeneous Face (LDHF) database consisting of 1 m indoor, and 60 m, 100 m, and 150 m outdoor VIS and NIR images of 100 different subjects is used for performing the tests of the proposed system.
Key Words: Cross Spectral, Cross Distance, Near Infrared (NIR), Visible Light(VIS), Long Distance Heterogeneous Face Database(LDHF-DB).
1. INTRODUCTION A primary modality for biometric authentication is FACE recognition and has received increasing interest in the recent years. Existing biometric systems are developed for cooperative user applications, such as access control, machine readable traveling document (MRTD), computer logon, and ATM. Such applications, requires a user to cooperate with the camera so as to have his/her face image captured properly. There are more general scenarios, such as face recognition for surveillance, where a person should be recognized without intentional or cooperative effort. To achieve reliable result, Face Recognition can be performed on intrinsic factors of face only. Among the several other factors, problems with uncontrolled environment lighting, is the issue to be solved for reliable face-based applications. A typical problem is that matching between near infrared (NIR) and visible light (VIS) face images in the situation that enrolment is finished with controlled indoor VIS face images, while authentication would be done using NIR face images to avoid the influence of the various environment illuminations. The sensing modality of NIR images is somewhat different than VIS images. These VIS images typically reside in face image databases usually maintained by the law enforcement agencies, such as the driver license databases, criminal records etc. Face recognition systems capable of performing
Š 2017, IRJET
|
Impact Factor value: 5.181
Fig -1: Visible light (VIS) & near-infrared (NIR) face images. Considering the degraded quality of image due to large distance and low illuminance during nighttime, a highquality probe face images is recovered from the degraded probe face images by using an Locally Linear Embedding (LLE) [ref] based image restoration method. Further a heterogeneous face matching [1] is used to match the restored face images to the database gallery face images.
2. BACKGROUND 2.1 NIR Images The capability of surveillance systems is expanded by different image acquisition methods. These methods collect
|
ISO 9001:2008 Certified Journal
|
Page 2799