Statistical Approach to a Colorbased Face Detection Algorithm

Statistical Approach to a Colorbased Face Detection Algorithm EE 368 Digital Image Processing Group 15 Carmen Ng, Thomas Pun May 30, 2002

Statistical Approach to a Color-based Face Detection Algorithm w Assumptions w 4 Stages: n n Pre-processing Skin Color Region Labeling Statistical Face Selection Techniques Edge Detection w Advantages/Disadvantages 2/15/2022 2

Statistical Approach to a Color-based Face Detection Algorithm Assumptions: w Color image w Multiple faces with similar area w Face orientation 2/15/2022 3

Statistical Approach to a Color-based Face Detection Algorithm I. Image Pre-processing w Boundary extension w Improves accuracy 2/15/2022 4

Statistical Approach to a Color-based Face Detection Algorithm II. Skin Color Region Labeling w Color-based n Chrominance extraction in YCb. Cr space w Morphological operations n 2/15/2022 Dilation and erosion 5

<= Original Image Rough Mask => 2/15/2022 6

Binary Mask after Morphological Operations 2/15/2022 7

Statistical Approach to a Color-based Face Detection Algorithm III. Statistical Analysis w Popular area finder n n Facial feature detector (holes in binary images) Popular area, width and height w Face rejection n 2/15/2022 Reject unpopular areas 8

Selected Face Regions after Stage III 2/15/2022 9

Statistical Approach to a Color-based Face Detection Algorithm IV. Facial Feature (Eye) Detection w Approximate eye location w LPF to remove noise w Edge detection to locate strong edges 2/15/2022 10

Typical Background Typical Face 2/15/2022 After LPF and Edge Detection 11

Statistical Approach to a Color-based Face Detection Algorithm Results/Conclusions: w 88% success rate w Adv: fast, no training required, work with video compression std. w Dis. Adv: min of faces required in image, work best with reliable facial detector 2/15/2022 12