Stereo matching for 2d face recognitionPresentation Transcript
DESIGN AND DEVELOPMENT FOR FACERECOGNITION USING STEREO MATCHING ALGORITHM by N.M.Harish BalajiSankara College of Science and Commerce
INTRODUCTION• Face Recognition(FR) - images & videos.• Face recognition compliments face detection .• Face detection - finds faces in images and videos .• Problems in FR - to handle pose variation .• 2 predominant methods 1) Geometric approach 2) Photometric approach
SECTIONS IN FACE RECOGNITIONFace Recognition deals with 3 main sections, they are: 1. Images with 3 landmarks in face. 2. Illumination variation. 3. Pose variation.
BRIEF PROCESS• FR handles pose & illumination variations.• Gallery image is generated with 4 landmark points.• Similarities are identified using matching cost.• Works well for large pose variations.• Dramatic changes is a challenging problem that an face recognition system needs to face.
FEATURES• Feature based system - detects - facial landmarks.• Initially face images need to be aligned. 1. To generate landmark points - Eyes, Nose, Mouth. 2. Fourth landmark - stereo.• Stereo - 3*3 filter - calculates the distance between test image & training image.
FACE RECOGNITION METHOD• Stereo Matching - supports good correspondence.• Dynamic programming - 2D face images.• Stereo algorithm - maximizes the cost function.
MATCHING PROCESS• Matching - individual pixel intensities.• Many matches .• Right match - difficult .
STEREO MATCHING• Stereo matching algorithm - individual pixel intensities.• Objective - Matching 2 images.• Matching - 2 scan lines l1 & l2.• Cost of matching is given by Matching cost = cost (l1,l2)