Top Banner

Click here to load reader

Face Categorization using SIFT features

Jan 02, 2016

ReportDownload

Documents

Face Categorization using SIFT features. Mei-Chen (Mei) Yeh ECE 281B 06/13/2006. Bush vs Schwarzenegger. Serena Williams vs Venus Williams. Main goal. Face categorization problem Object class recognition techniques have seen progress in recent years (ex: bag-of-words models) - PowerPoint PPT Presentation

  • Face Categorization using SIFT featuresMei-Chen (Mei) YehECE 281B06/13/2006

  • Bush vs Schwarzenegger

  • Serena Williams vs Venus Williams

  • Main goalFace categorization problemObject class recognition techniques have seen progress in recent years (ex: bag-of-words models)Different people are considered different object classesIntegration of SIFT featuresDoes SIFT help face recognition? To what degree?Target on a few peopleApplicationsFace annotation in family albumsName-based photo search

  • MethodFeatures: SIFTBag-of-words representation for facesEach SIFT feature is considered a codewordBuild a dictionary based on training samplesEach face is represented as a histogram over codewordsLearning: Nave Bayes Classifiers

  • SIFT feature vectors from training samples128-d feature spacecodeword 1codeword 2codeword 3

  • DatasetsThe BioID Face Database (simple)1521 images with 23 peopleVariety of illumination, background and face size

  • 22 categories, 1613 images70% for training, 30% for testing

  • MeasurementConfusion MatrixClassifiersCategoriesAverage Categorization Rate

  • Average: 89.14%

    1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 bg

  • DatasetsFaces in the Wild (challenging)851 images, 10 people + 1 non-facesExtracted from news videoshttp://www.cs.berkeley.edu/~millert/faces/faceDict/NIPSdict/

  • 11 categories, 851 images70% for training, 30% for testing

  • 81.82% 0.00% 0.00% 0.00% 0.00% 4.55% 4.55% 0.00% 4.55% 4.55% 0.00% 0.00% 83.33% 0.00% 0.00% 0.00% 0.00% 8.33% 8.33% 0.00% 0.00% 0.00% 4.00% 4.00% 68.00% 0.00% 4.00% 16.00% 4.00% 0.00% 0.00% 0.00% 0.00% 20.00% 0.00% 0.00% 80.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 8.00% 0.00% 0.00% 8.00% 68.00% 12.00% 0.00% 4.00% 0.00% 0.00% 0.00% 0.00% 0.00% 7.14% 0.00% 0.00% 85.71% 0.00% 7.14% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 100.00% 0.00% 0.00% 0.00% 0.00% 0.00% 9.09% 9.09% 0.00% 0.00% 4.55% 0.00% 77.27% 0.00% 0.00% 0.00% 0.00% 4.76% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 85.71% 9.52% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 8.33% 8.33% 83.33% 0.00% 0.00% 2.44% 2.44% 2.44% 2.44% 0.00% 0.00% 0.00% 2.44% 0.00% 87.80% Average: 81.91%

  • ConclusionsSIFT features + bag-of-words representation might work for face recognitionSimple dataset: goodChallenging dataset: may be improvedConsider the spatial relations between features may be the next step to improve the performance