03-07-2014, 03:29 PM
ARTIFICIAL INTELIGENCE FORFACE RECOGNITION
ARTIFICIAL INTELIGENCE FORFACE.docx (Size: 971.8 KB / Downloads: 11)
ABSTRACT
Today, Internet rules the world. The Internet is used to access the
complete facility of transferring the information, besides maintaining the
secrecy of the document. Since the network is considered to be insecure, the
encryption and authentication are used to protect the data while it is being
transmitted. The security is insufficient when the codes for encryption and decryption are revealed. There comes the necessity of increasing the security through face recognition usingneural network. Though it is costlier, it provides the high advantage of tight security. This paper deals with the recognition of images using neural networks. It is used in identifyingparticular people in real time or allows access to a group of people and denies access to the rest.The system combines local image sampling, the self-organizing map neural network,and a convolutional neural network. The self-organizing map provides the quantization ofimage samples into a topological space where inputs that are nearby in the original space arealso in the output space, thereby providing dimensionality reduction and invariance to minorchanges in the image sample. All these features are implemented using MATLAB v 6.5. Theconvolutional neural network provides for the partial invariance to translational, rotation,scale, and deformation. Hence it is analyzed that by implementing face recognition insecurity systems, the business transaction via Internet can be improved.
NOTE : The Matlab Codes will be shown at the time of presentation.
INTRODUCTION
The paper presents a hybrid neural network solution, which compares favorably withother methods and recognizes a person within a large database of faces. These neuralsystems typically return a list of most likely people in the database. Often only one image isavailable per person.First a database is created, which contains images of various persons. In the nextstage, the available images are trained and stored in the database. Finally it classifies theauthorized person’s face, which is used in security monitoring system. Faces representcomplex, multidimensional, meaningful visual stimuli and developing a computational modelfor face recognition is difficult.Face has certain distinguishable landmarks that are the peaks and valleys that sum upthe different facial features. There are about 80 peaks and valleys on a human face. Thefollowing are a few of the peaks and valleys that are measured by the software:
Distance between eyes
Width of nose
Depth of eye sockets
Cheekbones
Jaw line
Chin
These peaks and valleys are measured to give a numerical code, a string of numbers, whichrepresents the face in a database. This code is called a face print. Here the detecting,capturing and storing faces by the system is dealt with. Below is the basic process that couldbe used by the system to capture and compare images:
DETECTION
When the system is attached to a video surveillance system, the Recognition softwaresearches the field of view of a video camera for faces. Once the face is in view, it is detectedwithin a fraction of a second. A multi-scale algorithm, which is a program that provides a setof instructions to accomplish a specific task, is used to search for faces in low resolution. .The system switches to a high-resolution search only after a head-like shape is detected
ALIGNMENT
Once a face is detected, the head's position, size and pose is the first thing that is
determined. A face needs to be turned at least 35 degrees toward the camera for the system toregister it.
MATCHING
The newly acquired facial data is compared to the stored data and (ideally) linked toat least one stored facial representation. Briefly, the use of local image sampling and atechnique for partial lighting invariance, a self-organizing map (SOM) for projection of theimage sample representation into a quantized lower dimensional space, the Karhunen Loève(KL) transform for comparison with the self-organizing map, a convolutional network (CN)for partial translation and deformation invariance, and a multi-layer perceptron (MLP) forcomparison with the convolutional network is explored
LOCAL IMAGE SAMPLING
We have evaluated two different methods of representing local image samples. In
each method a window is scanned over the image as shown in figure .
1. The first method simply creates a vector from a local window on the image using the intensity values at each point in the window. Let xij be the intensity at the ith column, and the jth row of the given image. If the local window is a square of sides 2W+1 long, centered on xij, then the vector associated with this window is simply
[xi-W,j-W, W, W+1,..., ,..., i j ij x x xi+W,j+W-1, xi+W,j+W].
2. The second method creates a representation of the local sample by forming a vector out of
a) the intensity of the center pixel xij, and b) the difference in intensity between the center pixel and all other pixels within the square window. The vector is given by [xij-xi-W,j-W, W,W+1,..., ,..., ij i j ij ij x x wx xij-xi+W,j+W-1, xij-xi+W,j+W].
The resulting representation becomespartially invariant to variations in intensity of the complete sample. The degree of invariancecan be modified by adjusting the weight wij connected to the central intensity component.A depiction of the local image sampling process. A window is stepped over the image and a vector is created at each location
THE SELF-ORGANIZING MAP
Maps are an important part of both natural and artificial neural information
processing systems. Examples of maps in the nervous system are retinotopic maps in thevisual cortex, tonotopic maps in the auditory cortex, and maps from the skin onto thesomatosensoric Cortex. The self-organizing map, or SOM, is an unsupervised learningprocess, which learns the distribution of a set of patterns without any class information. Apattern is projected from an input space to a position in the map - information is coded as thelocation of an activated node. The SOM is unlike most classification or clustering techniquesin that it provides a topological ordering of the classes. Similarity in inputpatterns ispreserved in the output of the process. The topological preservation of the SOM process makes it especially useful in the classification of data, which includes a large number of
classes.The SOM is mainly used to find patterns in and classify high dimensional data,although it works equally as well with low dimensional data. The basic SOM consists of a 2-dimensional lattice L of neurons. Each neuron ni L has an associated codebook vector μi Rn . In what follows n , although in other applications n is often much larger. The lattice iseither rectangular or hexagonal with the connections within L determining the neighbourhood of a given neuron is shown in figure. Training the SOM involves first randomly initialising
all the codebook vectors and then sequentially presenting each training sample. A metric isfirst fixed on L, usually 2 other training algorithms exist, Euclidean or Manhattan.The SOM Lattice. Lattices are either rectangular or hexagonal; this in turn determines howmany neurons lie in each neighbourhood. All skin extraction experiments have used ahexagonal lattice
IMPROVING THE BASIC SOM
The original self-organizing map is computationally expensive due to:
1. In the early stages of learning, many nodes are adjusted in a correlated manner. Luttrel
proposed a method, which is used here, that starts by learning in a small network, and
doubles the size of the network periodically during training. When doubling, new nodes are
inserted between the current nodes. The weights of the new nodes are set equal to the average
of the weights of the immediately neighboring nodes.
2. Each learning pass requires computation of the distance of the current sample to all nodes
in the network, which is O (N). However, this may be reduced to O (log N ) using a hierarchy
of networks which is created from the above node doubling strategy
IMPLEMENTATION
The images in the training set; a fixed size window 5*5 is stepped over the entire image as
shown and local image samples are extracted at each step. At each step the window is moved
by 4 pixels. A self-organizing map is trained on the vectors from the previous stage. The
SOM quantizes the 25-dimensional input vectors into 125 topologically ordered values. The
three dimensions of the SOM can be thought of as three features. The same window as in the
first step is stepped over all of the images in the training and test sets. The local image
samples are passed through the SOM at each step, thereby creating new training and test sets
in the output space created by the selforganizing map. A Convolutional neural network is
CONCLUSION
The most talked about network, the internet, had thrown light on many unexplored
avenues. Security is one such avenue that forms the thumb rule for every application on the
internet. The technology to recognize human faces is implemented using neural network and
the thumbnails are successfully recognized thus proving it to be one of the best approaches
till date