Segmentation of Facial Images
Funded by NSF
Fig. 1: (a) A color image showing a face in a complex background. (b) Image obtained after mapping colors into intensities in such a way that the intensity at a pixel is proportional to the CIE Lab color distance of pixel to average skin color.
Fig. 2: The enhance operator, sigma equals 2.
Fig. 3: Transforming intensities of Fig.1b according to
the enhance operator.
Fig. 4: Smoothing of Fig. 3 with a 2-D Gaussian kernel. (a) Smoothing with a 2-D Gaussian kernel of standard deviation 2 pixels. (b) Smoothing with a 2-D Gaussian kernel of standard deviation 3 pixels. (c) Gradient magnitudes of (a) obtained by the Roberts operator. (d) Gradient magnitudes of (b) obtained by the Roberts operator.
Fig. 5: Single thresholding results of the smoothed face image shown in Fig. 4a. (a) Segmentation with T=26. (b) Segmenation with T=109. (c) Segmentation with T=162.
Fig. 6: (a)-(d) Thresholding the image of Fig. 4a at intensities equal to the average intensities of 3%, 6%, 10% and 15% highest gradient pixels in the image, respectively.
Fig. 7: Thresholding image of Fig. 4a with threshold values T1, T2, and the corresponding mask obtained by combining the segmentation results of T1 and T2. (a), (b) Threshold values are T1=56 and T2=147, respectively. T1 and T2 were obtained by setting parameter d=3. (c) Mask obtained by combining images (a) and (b). (d), (e) Threshold values are T1=26 and T2=162, respectively. T1 and T2 were obtained by setting parameter d=5. (f) Mask obtained by combining images (d) and (e). (g), (h) Threshold values are T1=12 and T2=174, respectively. T1 and T2 were obtained by setting parameter d=7. (i) Mask obtained by combining images (g) and (h).
Fig. 8: (a) Edges of Fig. 4a representing locally maximum gradients. (b) Edges of (a) corresponding to the top 10% highest gradients in the image. (c) Mask obtained by combining segmentation results of T1=26 and T2=162. (d) Overlaying (b) and (c). (e) Edges of (b) falling in mask (c).
Fig. 9: (a) Segmentation result using Fig. 4 with T=109. (b) Boundary of (a). (c) Edges obtained from Fig. 8e. (d) Searching closest edges along normal directions. Only some of the normals are shown here.
Fig. 10: Final edges obtained by expanding or shrinking the initial contour.
Fig. 11: (a) Fitting a rational Gaussian curve to points in Fig. 10 with Sigma=2. (b) Final segmentation result.
Fig. 12: (a) Fitting a rational Gaussian curve to points in Fig. 10 with Sigma=3. (b) Final segmentation result.
Initial results of face detection using a chroma chart.