Face is the most prominentfeature used to recognize any person. It is tough task to identify human facethrough facial expressions . Face recognition method using mathematic algorithmis the most competent system for the security systems. It is a biometrictechnology based concept with a purpose to prevent digital offences such as hackingof ATM machines, avoid voter’s fraud, suspect’s identification, human’s digitalinteraction with computer or other electronic machines, etc. Based on itsstructural algorithm, facial biometric system used prominent facial features toidentification.
Extorted facial features from face images are considerable forface biometrics system performance. Face recognition is arising hurdlein the field of computer visualization and pictorial analysis as such has acknowledgeda great deal of consideration over the last decade because of its manyapplications in several domains. It is the most hot topic since the late 80’s, deliveringsolutions to several realistic troubles. Face recognition method should be accurateto scan and identify face in images. when scanned result occur a recheck ofmany faces in a database using the various features.
The environment conditions like lighting,prospective view of a face, facial expressions and ageing. The similarity infaces makes it hard to recognize the exact face from database. This involvesextraction of its features from the image and then recognizes it, regardless ofpose, expression, occlusion, lighting, ageing and illumination. Face reorganization is notthat complex as it sounds, because it is easy to understand.
However, theprocess of recognition of faces used by the human brain foridentification has many factors yet to discover. Human brain sees the faces ondaily bases and number of faces he sees is very large. Brain uses the sameprocess for every face. Brain extracts the features from the faces and storethat features in his database, next time when he sees the same face, brain tryto recognize the face by extracting the feature and compare it the storedfeatures in his database.After comparing both the face, human brain tells theresult. This task is easy for human brain.
In modern era, digital security andbiometric based authority identification is in demand for banks, offices,confidential research holds etc. Face Recognition Technology is used inidentity authentication, access control and other highly featured applications.In present time, still there is more detailed research is conducting forimprovements and reducing flaws of biometric technology. Basically, facerecognition simply works with face scanned images, face viewpoint and otherpre-set expressions. Therefore, the functionality of the face recognitionsystem is always a matter of doubt.
In this thesis, twoalgorithms are used I,e. Local Binary Pattern and Uniform Local Binary Pattern.Local Binary Pattern (LBP) was first defined in 1994.
It is a type of featureused for the purpose of classification in computer vision. It has been observedthat LBP produces the best classifier of humans when it is combined with Histogramof oriented gradients (HOG) classifier. Uniform Local binary pattern is said tobe uniform if upon traversing the bit pattern in a circular way, the binarypattern contains at most two bitwise transitions from 0 to 1 or vice versa. Thesepatterns are used in the computation of LBP labels so that there is a separatelabel for each uniform pattern and but a single label for all the non-uniformpatterns. In this proposed work, wehave studied the work on both i.e. face recognition and detection techniquesand developed algorithms for them. We have used a novel LBP-block wise facerecognition algorithm based on the features of local binary pattern anduniforms local binary pattern.
Two variants (each) of both the techniques areused in this thesis work. One variant is applied to whole image as it is and inother variant, first image is divided into 16 blocks of size 4 X 4 and then thetechniques are applied to each block and in the end the features of all theblocks are appended together to form a combined feature vector for the wholeimage. We have validated the proposed method on a set of classifiers computedon a benchmarked ORL image database. It has been observed that the recognitionrate of the novel LBP-block wise method is better than the existing LBP method.
Place: Ferozepur Nikhil Kumar Date: January 11, 2018