computeFaceAlignmentTransform
fun computeFaceAlignmentTransform(rightEye: FacePoint2D, leftEye: FacePoint2D, noseTip: FacePoint2D, mouthCenter: FacePoint2D, targetSize: Int): SimilarityTransform2D?
Computes the optimal 2D similarity transform (Umeyama / Procrustes least-squares alignment) from the 4 central BlazeFace landmarks (rightEye, leftEye, noseTip, mouthCenter) to a canonical square target face template of size targetSize x targetSize.
Unlike 2-point eye-only alignment, anchoring across the eyes, nose tip, and mouth center stabilizes crop scale and vertical positioning when eyewear frames or lens refraction perturb the detected eye keypoints.
Returns null if the landmarks are degenerate (collinear or collapsed).