DESCRIPTION
The patent disclosure covers System and Method for Deep Learning Based Attendance Monitoring System Using Facial
Recognition.
This patent disclosure the system is instantiated by the mobile. After it triggers then the system starts processing the image
of the students for which we want to mark the attendance. Image Capturing phase is one in which we capture the image of
the students. This is the very basic phase from which we· start initializing our system. We capture an image from our camera
which predominantly checks for certain constraints like lightning, spacing, density, facial expressions etc. The captured image
is resolute according to our requirements. Once it is resolute, we make sure it is either in .png or .jpeg format. We take different
frontal postures of an individual so that the accuracy can be attained to the maximum extent. This is the training database in
which we classify every individual based on labels. For the captured image, from an every object we detect only frontal faces.
This detects only face and removes every other part since we are exploring the features of faces only. These detected faces
are stored somewhere in the database for further enquiry. Features are extracted in the extraction phase
evaluation and benchmarking of these algorithms is cruciaL One of the most important facts learned in ·
these evaluations is that large sets of test images are essential for adequate evaluation. It is also extremely
important that the samples be statistically as similar as possible to the images that arise in the application
being considered. Scoring should be done in a way that reflects the costs of errors in recognition. In
planning an evaluation, it is important to keep in mind that the operation of a pattern recognition system is
statistical, with measurable distributions of success and failure. These distributions are very applicationdependent
, and no theory seems to exist that can predict them for new applications. This strongly suggests
that an evaluation should be based as closely as possible on a specific application.
The characteristics of faces/human body parts. During the past eight years, research on human
action/behavior recognition from video has been very active and fruitfuL Generic description of human
behavior not particular to an individual is an interesting and useful concept. One of the main reasons for
the feasibility of generic descriptions of human behavior is that the intraclass variations of human bodies,
and in particular faces, is much smaller than· the difference between the objects inside and ?utside the
class. For the same reason, recognition of individuals within the class is difficult. For example, detecting
and localizing faces is typically much easier than recognizing a specific face. Before we examine existing
video-based face recognition algorithms, we briefly review three closely related techniques: face
segmentation and pose estimation, face tracking, and face modelling. These techniques are critical for the
realization of the full potential of video-based face recognition .
CLAIMS
IIWe Claim,
1. The patent disclosure covers System and Method for Deep Learning Based Attendance Monitoring
System Using Facial Recognition as described above in Fig 1 and 2.
2. In this various algorithms for the implementation of face recognition system in mobile phones. Eigen
faces machine-learning algorithm was the engine of training the system after applying some filters
on the image.
3. Furthermore, the Eigen faces algorithm allows the application to recognize the face Realtime. Eigen
faces was not very sensitive )o a change in the number of subjects during the first phase, however,
an increase in size of the training set helped the algorithm to correct its wrong prediction.
4. An increase in the data set did not help to recognize more subjects, but it turned correct predictions
into wrong ones. Eigen face was not accurate in the second phase. Fisher faces had better results
phase. Sometimes with 20 pictures, the results were better, but with 40 pictures the results were the
same or worse.
5. This algorithm was most of the time the worst in the two phases. It also had a very low accuracy in
the second phase. A face is detected using the Local binary pattern cascade classifier. After testing
the Haar-like cascade classifier, the speed of the detection was very low as compared to LBP that
always had at most 96 % in the first phase and is comparatively better algorithm.
6. In the second phase, this algorithm had a.consequent drop in its accuracy compared to the first
phase. An increase in the number of ubjects dramatically changed its prediction. In each phase, an
increase in the training data had a positive effect or no effect.
7. As part of the future work, we would like to develop an application that would allow the user to add
or delete face classes in the training set. This would give users the freedom to define their own user
groups rather than a pre-defined set on the server.
8. We would also like to explore better algorithms for face detection and face recognition to increase
the number of student to be detected and recognize