吳家宇 吳明翰 Face Detection Based on Template Matching and 2DPCA Algorithm 2009/01/14.

吳家宇吳明翰

Face Detection Based on Template Matching and 2DPCA Algorithm

Face Detection Based on Template Matching and 2DPCA Algorithm

2009/01/14

IntroductionIntroductionIntroductionIntroduction• Face recognition

• Face detection• Feature extraction• Feature matching

• Related Methods• Skin-based detection method• Neural networks method• SVM

• In this paper• Template matching algorithm• 2DPCA algorithm

Face DetectionFace DetectionFace DetectionFace Detection

• First step• Preprocess the image

• Normalize• Histogram equalization• Luminance compensation

• Second step• Search rectangle regions

• Last step • Detect every detection window hierarchically

Face Detection AlgorithmFace Detection AlgorithmFace Detection AlgorithmFace Detection Algorithm

FlowchartFlowchartFlowchartFlowchart

• Resize the image– Resized into 400*300 pixel

• Histogram equalize the image

Pre-ProcessingPre-ProcessingPre-ProcessingPre-Processing

• Two-eye template• Choose 30 standard face images and cut out a pair of eyes regions• Get a 20*10 pixel two-eye template via calculating average number to many couples of eyes

regions

• Face template• Enlarged nearby based on the eyes template and construct 20*25 pixel face template• Most non-face image blocks which interrelated coefficient is less than value T are discarded

Template MatchingTemplate MatchingTemplate MatchingTemplate Matching

Two-eye template Face template

•Denotes separately •a gray matrix •average value•Standard deviation

• PCA– Matrices-to-vector conversion

• High dimensional vector space• Difficult to evaluate covariance matrix• Time-consuming

• 2DPCA– Directly computes eigenvectors of image covariance matrix– More efficiently than PCA– Easier to evaluate covariance matrix– Less time to determine the corresponding eigenvectors

Comparison between PCA and 2DPCAComparison between PCA and 2DPCA

2DPCA2DPCA2DPCA2DPCA

2DPCA2DPCA2DPCA2DPCA

• Minimal distance classifier– Realizes classification to every image matrix– Let– The detection windows corresponding to B are taken account of face region– Otherwise the detection windows are non-face

Classification-MergingClassification-Merging

• After two hierarchical detection– Most faces may be detected at multiple nearby positions or scales– Overlapping detected windows should be merged

• Merging

ComparisonComparisonComparisonComparison

ResultResultResultResult

ResultResultResultResult

Training DataTraining DataTraining DataTraining Data

• Bao Face Database – Lots of face images, mostly people from Asia. Single face

pictures are in the "one faces" subdirectory.

DemoDemoDemoDemo

DemoDemoDemoDemo