1) The document proposes a real-time unconstrained face recognition system using deep convolutional neural networks (DCNN). 2) The system performs face detection, extracts DCNN features, and computes similarity to perform face recognition on images and video frames. 3) It was tested on challenging datasets like CASIA-WebFace, IJB-A, and LFW and was able to achieve accurate recognition with variations in pose, illumination, expression, resolution and occlusion.

1. A DCNN APPROACH FOR REAL-TIME
UNCONSTRAINED FACE RECOGNITION

2. INTRODUCTION
The problem of understanding and tracking a non-rigid object that has unpredictable differences
in shape and appearance (for example, a human face) is difficult, and the research in this domain
has developed many efficient and accurate solutions. Many algorithms have been used to work
well on images that are collected in controlled settings. However, the performance of these
algorithms often degrades the quality of images that have large variations in color, lighting,
expression, aging, and smoothness. To solve this problem, many methods have focused on
learning invariant and discriminative representation from face images and videos. Using local
photometric features for face recognition in the more complex canvas has become a widely
accepted method. In that setting, the typical method is detecting interest points or regions in
input images.

3. METHOD
Our system is a real time unconstrained face recognition. Face detection is done to localize and align faces in each image and video frames. Next train DCNN
using CASIA-WebFace and derive joint bayesian metric using IJB-A dataset and DCNN features. Finally we perform face recognition by computing similarity.
Fig 1.gives the block diagram of our system. The working model of our system is illustrated in Fig 2.The details of each component of our approach are
presented in the following sections
• Face detection and processing
All the images/frames are converted into grayscale. To detect face, we use haar wavelets. We perform image compression using discrete cosine
transform which is a lossless image compression method. Then, each face is aligned into the canonical coordinate with similarity transform using the 5
landmark points ( i.e. two left eye corners, two right eye corners and nose tip). After alignment, the face image resolution is 25 x25 pixels.
• Feature Extraction
For feature extraction and dimensionality reduction we use two Algorithms principal component analysis and pose invariant face recognition.
PCA is an algorithm for reducing dimensionality of a feature space by projecting it on a space that spans the significant variations. It is an unsupervised linear
transformation technique. PCA helps to identify patterns in data based on the correlation between features.PCA finds the eigenvectors of a covariance matrix
with the highest eigenvalues and then uses those to project the data into a new subspace of equal or less dimensions. PCA converts a matrix of n features into a
new dataset of less than n features. That is, it reduces the number of features by constructing a new, smaller number variables which capture a signficant portion
of the information found in the original features. Extending the frontal view face recognition system to a pose-invariant recognition system is quite simple if
one of the proposed specifications of the face recognition system is relaxed. Successful pose-invariant recognition will be possible if many images of a known
individual are in the face database. Nine images from each known individual can be taken as shown below. Then if an image of the same individual is
submitted within a 30o angle from the frontal view he or she can be identified.

4. As per our study in the proposed domain, the existing methods are follows both manual and automated
methods. Now the biometric authentication systems are follows fingerprint scanning methods and RF ID tags. In
the fingerprint scanning system, the authentication processes with electricity, they use capacitors and optical
sensors to generate the sense. Also, this system stores the finger data at the chip level. The use of capacitors may
reduce the accuracy of the system. Sometimes we need to wipe the entire sensor for smooth performance, in this
case, we need an alternative method along with this system like a pin code.
RFID is being used actively in retail, healthcare, and other sectors to monitor workers. Since the workers in these
sectors are large in number, hard to handle and their work can be performed by others in case of absenteeism;
there the attendance mechanism is of trivial significance. The System is expensive because a lot of technology
goes into making it. In the case of a large strength of people, purchasing tags for everyone is costly. Replacing
the system's microchip, radio transceiver, antenna, and the battery is tiresome and costs money. Since it is not as
secure as biometrics, the system is prone to manipulation.
EXISTING SYSTEM

5. PROPOSED SYSTEM
In face verification, given videos or images, contain multiple face coordinates. The objective
is to identify the faces and name them. Deep convolutional neural networks (DCNN) are
introduced to perform different tasks such as object recognition, object detection, and face
verification. In this work, we train a DCNN model using a relatively small face dataset, the
CASIA-WebFace and compare the performance of our method with other commercial face
matches on the challenging IJB-A dataset which contains significant variations in pose,
illumination, expression, resolution, and occlusion. We also evaluate the performance of the
proposed method on the LFW dataset.

6. ADVANTAGE
 Time Saving
 Implementation Cost is low
 Accessibility
 Accuracy
 Speed

7. BLOCK DIAGRAM

8. CONCLUSION
In this system, we propose real-time unconstrained face recognition from both image and video
using DCNN features. The computational models, which were implemented in this project,
were chosen after extensive research, and the successful testing results confirm that the choices
made were reliable. This system was tested under very robust conditions in this experimental
study. We also study the performance of the proposed DCNN on the IJB-A dataset.