Facial Recognition of Human in A Real Time Video Using PCA Algorithm

—The human facial recognition has been attractive research area in computer vision. An existing facialrecognition system is based on still images, facing complex problem in discriminating foreground frombackground cluster without motion information. To overcome this problem the facial recognition invideo motion is projected. The execution work is based on haar cascading classification for facedetection and PCA (Principal Component Analysis) for Facial recognition. The database is trained with100 sample faces of 10 people with different poses, known as positive and negative example ofarbitrary image of the size 20 x20. The experimental results are compared with fisher face andLBP(Local binary pattern) but PCA provides 92.4% accuracy.


I. INTRODUCTION
In the field of computer vision, one of the most important and curial area is the facial recognition.The aim of facial recognition is to dynamically analyze and recognize human face from a video. The human recognition is one of the most powerful tool, that the human can identify the unique feature from their faces. Thus the recognition of face in human through facial recognition has become a great deal during the last decades. The ability to detect and recognize human faces from videos helps to monitor the behaviour of objects. The several applications based on surveillance, which are used in ATM centers, Railway stations, companies and so on. These applications are mainly used to prevent criminal activities, and it also helps to monitor state and behaviour of human. In recent studies, many algorithms are being focused on face recognition, age calculation and facial emotion classifications. In this context, we implemented face recognition system based on PCA algorithm, it consist of Reading real time video from webcam, Face Detection, Face Extraction, Face alignment and finally recognition and representation is done by using Database. Some of the drawbacks in face detection system is to detect the face in which the person wears glass, hat and having moustache on his face. The recognition work based on capturing a real time video using webcam and clipped into many frames. Each frame is referred by Haar cascade trained classifier. The Haar feature based classifier is an effective object detection method to detect the face from frame. The detected face is taken for face estimation. In this context, the feature alignment of detected face is constructed through the ratios of the facial features for instance nose, eyes, lips, and so on. The alignment value of detected face is used for recognisation, by comparing the detected face with the database image. The database is enclosed with set of human face with details like name. The details of the human face are saved by the training set and can also be deleted. This system can detect multi-faces. However it can recognize only one face at a time.
II. EXISTING SYSTEM An existing recognition system is based on still images, for recognition purpose the collection of still images are taken as input dataset. Each image is separately processed by eliminating background and foreground of an image. The designed descriptors for this facial recognition system are based on set of factors like relative positions, size and distance of faces for investigating human recognition 1. In current face detection algorithm provides less satisfactory for complex situations. For an example the view of face in different angle, illumination, occlusion, weather changes, facial appearance, shape, etc., at present the multi-view face detection is still quite challenging task [17]. Recognizing face from the video is difficult task, because detected face image is smaller than the frame size, for this the recognition needs a lot of improvement has to be made. For improvement various algorithms are used, which are Template Based approaches, Statistical approach, Model-based approaches, Wholistic approaches. The template Based approach is use an Adaptative Appearance Models. The Statistical approach of recognition algorithms are PCA (

CAPTURING VIDEO
The first phase of facial recognition system is to acquire real time video by using webcam, as an input video stream is used for the face detection process. The accuracy of the video depends on the distance of object. The reading video is processed by capturing a frame from the video.

FACE DETECTION
The face detection is a technique in computer vision. It is used to identify the human face in a real time digital video, which refers psychological process of locating and identifying human face for various applications which are surveillance, authentication, etc., In this context, each frame of real time video is converted into gray level image, then the genetic algorithm of Haar-Cascade Classifier is used to locate the face in a frame. The classifier is trained with few sample faces known as positive examples, they are assigned with a particular size 20 X 20 and negative examples will act as the arbitrary images of the same assigned size. After the classifier is trained, it can by allot to the desired region in an input image. If the result of classifier is '1', mean that the selected region is depicting to show the face or else it results as '0'.The design of the classifier is so efficient, that can be resized for better detection purpose. The detection region is processed by comparing the facial templates. The detection will depend on factors like distance, clarity, motion etc. Finally the detected face is highlighted using a rectangular box.
The key features of Haarcascade classifier are: Step 1: Calculate the Integral Image An Integral image is a process of calculating Summed Area Table (sum of values or pixel values). In this process every block is the summation of the previous blocks above it [2] as shown in the figure 2. Here the point of origin is the top left corner of the block and the previous blocks are those to the left of the as well as the above blocks. Its purpose is to allow quick computation of any area in an image with only four memory lookups and three addition or subtraction operations. These produces two equations, they are:  (From Eq.1) Step 2: Apply Haar-Features The Haar Feature is a digital features, it is used for recognition purpose. They are simple rectangular features that achieve random accuracy and are mainly used for their simplicity and fast computation time [2]. It is known as difference calculation. The difference between the entire whites block and the black blocks i.e, feature = Sum(White)-Sum(Black). There are numerous haar-like features available. Choose a subset of 5 more basic features. They are: Based on the integral features, the edge features havesix memory lookups (two points at bottom, two points in middle and two points at the top).The line features have eight memory lookups(two points at the bottom, four points in the middle and two points at the top).The four-rectangular features have nine memory lookups (three points at bottom, two points in middle, one point at the centre and three points at the top). These variables are programmed with adjustable height and width dimensions for scaling [2]. Step 3: Use Adaboost Learning Algorithm It stands for adaptive boosting and is essentially the training algorithm that selects the best features out of the enormous over complete set of features available and creates a strong classifier. This algorithm is an aggressive approach which disregard the majority of features and results a fewer features. This provides a weak classifier (Wj)which consists of Features (Fj), Threshold (Ɵj), Polarity (Pj).
The positive images are only trained. Set the threshold value based on the average pixel intensities and standard deviation. This helps in faster implementation of code and require less data.
Step 4:Apply Cascade Filter This process essentially discards all the negative sub-windows as earily as possible to reduce the over all computational time and focus more on possible positive windows. This is a binary tree created with all strong features from the adaboost method. This process allows possitive matches only to be revaluated at any point anything that registers its value as negative sub-window gets immediately rejected and exists. It reduces computation time spend on false windows by rejecting them. The threshold values can be adjusted to allow certain accuracy. Lower thresholds yeild higher detection rates but increase the occurance of false positives.

TRAINING SET CREATION
The extracted face of Haar-Cascade Classifier is dynamically stored in a database, which are used as training sets for recognition process. The training set contains details about the extracted face feature.

FACE RECOGNITION
Facial recognition is a biometric method, to identify an individual face by comparing live capture or digital image data with the database for that person. Facial recognition system is commonly used for security purposes but is increasingly being used in a variety of other applications. The facial recognition system is done by using Principal Component Analysis (PCA) algorithm. PCA is used for analyzing data (feature)and it is used to reduce the dimensionality of the feature. The major advantage of PCA is to reduce noise sensitivity, and increase efficiency. PCA facial recognition was performed by construction of Euclidean distance between feature vectors of detected face. PCA algorithm use covariance matrix and eigenvector for recognition. Eigenvector of the covariance matrix of the set of face image β (a,b), where an image with N x N pixel is considered a point in N2 dimensional space [7]. Step 1: Training process  Here the Training face are represented as T1,T2,T3,………TM with m images of size N x N are represented by vector size N2. Each face is represented by a variable β1, β2, β3,……….βm. The transferred face vector are placed into a training set Z. Z contains {β1, β2, β3,……….,βm} Step 2: Compute average face vector To calculate average face vector (δ) use the formula, δ = 1 n Step 3: Covariance matrix (C) Subtract average face vector from each face vector and stored it in a variable £Ito calculate Covariance matrix (C). £i = β1 -δ(5) C = S S T (N 2 x N 2 matrix) (6) Where S=[£1,£2,£3,………,£M] (N2 x M matrix).
Step 4: Calculate the eigenvector and eigen values of the covariance matrix The dimensionality of the covariance matrix is N2 x N2. Here the total number of eigenvector will be greater than the total number of training set. The final result of this derivation is that S ri is an eigenvector of C = SST. yi forms the eigenface of M eigenvector of C' = STS Step 5: Represent each face image a linear combination of all K eigenvector. Each face from training set can be represented a weighted some of the k eigenface and the mean face because the removed average face vector must be added. If percentage of eigenvector is multiplied with the total eigenvector the percentage of resemblance of the original face can be obtained.
IV. RESULTS AND DISCUSSION The facial recognition system of human in a real-time video using PCA algorithm was implemented in C#.Net. The result of PCA Algorithm is shown in the figure 9,10,11,12,13. Here sensitivity to variations in pose of face and illumination changes is still a challenging problem. The performance evaluation of the facial recognition system of identification rate (%) as shown in the Table 1.  The training set is created by assigning Authentication names to each detected face. The training set image is uploaded in the database for identification. The database contains 100 image, few of them are shown in figure  13. if any detected face match with the database image, it will display name at top the face else it will display unknown.
V. CONCLUSION The human face is detected and recognized by using Haar Cascade Classifier algorithm and PCA algorithm. The database is trained with few sample faces known as positive example and negative example which will act as the arbitrary image of the same size 20 x 20. The database contains 100 images of 10 people with different poses. Some features where not detected, like the person wearing glass, hat and having moustache on his face. Advantage of PCA is to reduce the dimensionality of features by using covariance matrix and eigenvector. Here the human faces are recognized with the training set where the images are stored in the database. This system helps to detect multiple faces. Comparative study is done with LBP and Fisher face algorithm, the experimental result of PCA provides 92.4% accuracy.