Key Landmarks Detection of Cleft Lip-Repaired Partially Occluded Facial Images for Aesthetics Outcome Assessment

Paul M. Bakaki, Bruce Richard, Ella Pereira, Aristides Tagalakis, Andy Ness, Ardhendu Behera, Yonghuai Liu

Abstract: This paper proposes a novel method for the detection of the symmetrical axis of the cropped face required for the aesthetic outcome estimation from the facial images of patients after their cleft treatment. It firstly applies the Gaussian filter to smooth the images in order to compress noise on the subsequent tasks, then the bilateral semantic segmentation network is applied to segment the facial components out and each region is assigned a distinct colour, thirdly the Canny edge detector is applied to detect the facial feature points and all the contours are further detected and classified into three thirds according to their height. Fourthly, the centres of mass of detected feature points on the contours and the average of all these centres are used to estimate four potential symmetrical axes of the face, the one with minimum Manhattan distance from all the detected feature points is finally selected as the optimal one and used to estimate the aesthetic numerical score through the shape analysis in structural similarity measure. The experimental results based on a publicly accessible dataset shows that it performs well and better than one existing method.