1 / 12

EE368 Face Detection Project Report Group 6

EE368 Face Detection Project Report Group 6. 04880489 Yuichiro Yamashita yamas@stanford.edu. Contents. Algorithm selection System design Results Conclusion. Algorithm Selection. Main idea: Support Vector Machine Classifier using samples which lie on the decision boundary

kynton
Télécharger la présentation

EE368 Face Detection Project Report Group 6

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. EE368 Face Detection Project ReportGroup 6 04880489 Yuichiro Yamashita yamas@stanford.edu

  2. Contents • Algorithm selection • System design • Results • Conclusion

  3. Algorithm Selection • Main idea: Support Vector Machine • Classifier using samples which lie on the decision boundary • Problem: Slow on Matlab • Prefilter before SVM • non-face rejection by color: using HSV Color space for skin color detection • non-face rejection by pattern: using Maximal Rejection Classifier (MRC)

  4. Recipe • Apply Skin Color Detector for Image • Create Image pyramid and apply MRU. • Take AND operation of 1 and 2, then classify by SVM • Merge Image pyramid, and remove repetitive face candidates.

  5. Skin Color Detector • Avoid false rejection • Faster is better • No need to be precise • The pixel with Hue of -330deg to 45deg is regarded as face region. Skin pixels plotted in HS-space. Hue being the angle q Work from J. Sherrah and S. Gong

  6. Prefilter by MRC • Train MRC by faces and non-faces of 19 x 10 block size (Upper half of the face) • Training set contains 543 faces and 477 non-face blocks manually picked-up and 100000 arbitrary picked-up non-faces images • Important: Avoid False Rejection

  7. Support Vector Machine • Train SVM by faces and non-faces of 19 x 19 block size. • Training set contains 543 faces and 293 non-face blocks. • Horizontal and Vertical differentiation to cancel DC offset. • Main Ideas from Papageorgiou, 1998 • SVM toolbox v0.54 by Cawley.

  8. Skin Color Detection: Result • Filtering by Hue value • Walls etc. are still classified as skin

  9. MRC: Result • Blight pixels are candidates for face. • Still some false positives found

  10. Face Detection Result after SVM

  11. Face Detection: Result • Images 1, 2, 3, 6, 7 are used for training, 4,5 is for test only • Detection rate of 86%, with 1 false alarm. • Can be improved with increased number of layers and giving more face examples.

  12. Conclusion • Face Detection System has been implemented with Support Vector Machine • Pre-filter by Skin Color and Maximal Rejection Classifier has been used to expedite the process • Reasonable Detection Rate. Can be easily improved with increased number of layers and face/non-face samples

More Related