Facial Scan Change Detection

We present a method for quantifying and localising changes in two facial scans of the same person taken at two different time instants. The method is based on rigid registration and semantic feature extraction, followed by discrepancy computation. The proposed method combines the Landmark Transform (LT) method, which is applied on semantic feature points, and the Iterative Closest Point (ICP) algorithm, which is performed on semantic regions. Finally, the discrepancy between the two scans is computed using the Symmetric Hausdorff distance. Experimental results with both synthetic and real data show the effectiveness of the proposed method which has also been validated by an experienced clinical scientist. Moreover, the method is being used as support in clinical studies on a 3D object database with more than 1000 facial scans.