Preprocessing Images for Facial RecognitionSolutionProcessMethods to TestCalculating EigenfacesVerifying FacesExample EigenfacesResultsReferencesPreprocessing Images Preprocessing Images for Facial Recognitionfor Facial RecognitionAdam SchreinerAdam SchreinerECE533ECE533SolutionSolutionFace recognition systems have problems Face recognition systems have problems recognizing differences in lighting, pose, recognizing differences in lighting, pose, facial expressions, and picture quality.facial expressions, and picture quality.ProblemProblemApply some sort of processing to Apply some sort of processing to images before they are analyzed to images before they are analyzed to increase successincrease successProcessProcessCreate a training setCreate a training setRead in imagesRead in imagesApply Preprocessing techniqueApply Preprocessing techniqueMake it easier to process the data and increase the chances Make it easier to process the data and increase the chances of getting correct matchesof getting correct matchesBetter chances of success with changes in illumination, pose, Better chances of success with changes in illumination, pose, picture quality.picture quality.Decrease processing time.Decrease processing time.Format data, calculate the face spaceFormat data, calculate the face spaceApply same Preprocessing technique to test imagesApply same Preprocessing technique to test imagesRun test images against the face spaceRun test images against the face spaceRank techniques based on number of correct matches, number Rank techniques based on number of correct matches, number of false matches, and time to calculate dataof false matches, and time to calculate dataMethods to TestMethods to TestSmoothingSmoothingBlurringBlurringSharpenSharpenEdge DetectionEdge DetectionImage SizeImage SizeCombinationsCombinationsCalculating EigenfacesCalculating EigenfacesRead in Training Set Read in Training Set Apply Processing Technique Apply Processing Technique Calculate the mean imageCalculate the mean imageFind the difference between Find the difference between each image and the mean each image and the mean imageimageCalculate L matrix and eigen Calculate L matrix and eigen vectors vectors Calculate eigenfacesCalculate eigenfacesKeep the M’ images that Keep the M’ images that correspond to highest eigen correspond to highest eigen values as the face spacevalues as the face spaceiiMnn1M ,....,21],...,[ 21 MA )(iMkklkiu1AALTVerifying FacesVerifying FacesForm a set of weights from Form a set of weights from training datatraining dataGet new face image, apply Get new face image, apply preprocessing techniquepreprocessing techniqueFrom set of weights for new From set of weights for new imageimageFind the distance between the Find the distance between the new face and the training datanew face and the training dataIf distance is less than a set If distance is less than a set threshold the face is threshold the face is categorized as the kcategorized as the kthth person person in the database.in the database.)( lTkku],...,[ '21 MTk)( Tkku22kk],...,['21 MTExample EigenfacesExample EigenfacesResultsResultsWill be ranked on best performance based uponWill be ranked on best performance based uponCorrect matchingCorrect matchingSpeedSpeedIncorrect matchesIncorrect matchesReferencesReferencesM. Turk, A. Pentland, Eigenfaces for Recognition, Journal of M. Turk, A. Pentland, Eigenfaces for Recognition, Journal of Cognitive Neurosicence, Vol. 3, No. 1, 1991, pp. 71-86 Cognitive Neurosicence, Vol. 3, No. 1, 1991, pp. 71-86 W. Zhao, R. Chellappa, A. Rosenfeld, P.J. Phillips, Face W. Zhao, R. Chellappa, A. Rosenfeld, P.J. Phillips, Face Recognition: A Literature Survey, ACM Computing Surveys, 2003, Recognition: A Literature Survey, ACM Computing Surveys, 2003, pp. 399-458 pp. 399-458 WikipediaWikipediahttp://ai.ucsd.edu/Tutorial/matlab.htmlhttp://ai.ucsd.edu/Tutorial/matlab.htmlECE533 Course NotesECE533 Course NotesECE738 Course NotesECE738 Course
View Full Document