華為雲——AI人臉編輯讓Lena微笑

import numpy as np

import PIL

import PIL.Image

import scipy

import scipy.ndimage

import dlib

def get_landmark(filepath, predictor):

     """get landmark with dlib

     :return: np.array shape=(68, 2)

     """

     detector = dlib.get_frontal_face_detector()

     img = dlib.load_rgb_image(filepath)

     dets = detector(img, 1)

     for k, d in enumerate(dets):

         shape = predictor(img, d)

     t = list(shape.parts())

     a = []

     for tt in t:

         a.append([tt.x, tt.y])

     lm = np.array(a)

     return lm

def align_face(filepath, predictor):

     """

     :param filepath: str

     :return: PIL Image

     """

     lm = get_landmark(filepath, predictor)

     lm_chin = lm[0: 17]  # left-right

     lm_eyebrow_left = lm[17: 22]  # left-right

     lm_eyebrow_right = lm[22: 27]  # left-right

     lm_nose = lm[27: 31]  # top-down

     lm_nostrils = lm[31: 36]  # top-down

     lm_eye_left = lm[36: 42]  # left-clockwise

     lm_eye_right = lm[42: 48]  # left-clockwise

     lm_mouth_outer = lm[48: 60]  # left-clockwise

     lm_mouth_inner = lm[60: 68]  # left-clockwise

     # Calculate auxiliary vectors.

     eye_left = np.mean(lm_eye_left, axis=0)

     eye_right = np.mean(lm_eye_right, axis=0)

     eye_avg = (eye_left + eye_right) * 0.5

     eye_to_eye = eye_right - eye_left

     mouth_left = lm_mouth_outer[0]

     mouth_right = lm_mouth_outer[6]

     mouth_avg = (mouth_left + mouth_right) * 0.5

     eye_to_mouth = mouth_avg - eye_avg

     # Choose oriented crop rectangle.

     x = eye_to_eye - np.flipud(eye_to_mouth) * [-1, 1]

     x /= np.hypot(*x)

     x *= max(np.hypot(*eye_to_eye) * 2.0, np.hypot(*eye_to_mouth) * 1.8)

     y = np.flipud(x) * [-1, 1]

     c = eye_avg + eye_to_mouth * 0.1

     quad = np.stack([c - x - y, c - x + y, c + x + y, c + x - y])

     qsize = np.hypot(*x) * 2

     # read image

     img = PIL.Image.open(filepath)

     output_size = 256

     transform_size = 256

     enable_padding = True

     # Shrink.

     shrink = int(np.floor(qsize / output_size * 0.5))

     if shrink > 1:

         rsize = (int(np.rint(float(img.size[0]) / shrink)), int(np.rint(float(img.size[1]) / shrink)))

         img = img.resize(rsize, PIL.Image.ANTIALIAS)

         quad /= shrink

         qsize /= shrink

     # Crop.

     border = max(int(np.rint(qsize * 0.1)), 3)

     crop = (int(np.floor(min(quad[:, 0]))), int(np.floor(min(quad[:, 1]))), int(np.ceil(max(quad[:, 0]))),

             int(np.ceil(max(quad[:, 1]))))

     crop = (max(crop[0] - border, 0), max(crop[1] - border, 0), min(crop[2] + border, img.size[0]),

             min(crop[3] + border, img.size[1]))

     if crop[2] - crop[0] < img.size[0] or crop[3] - crop[1] < img.size[1]:

         img = img.crop(crop)

         quad -= crop[0:2]

     # Pad.

     pad = (int(np.floor(min(quad[:, 0]))), int(np.floor(min(quad[:, 1]))), int(np.ceil(max(quad[:, 0]))),

            int(np.ceil(max(quad[:, 1]))))

     pad = (max(-pad[0] + border, 0), max(-pad[1] + border, 0), max(pad[2] - img.size[0] + border, 0),

            max(pad[3] - img.size[1] + border, 0))

     if enable_padding and max(pad) > border - 4:

         pad =外匯跟單gendan5.com np.maximum(pad, int(np.rint(qsize * 0.3)))

         img = np.pad(np.float32(img), ((pad[1], pad[3]), (pad[0], pad[2]), (0, 0)), 'reflect')

         h, w, _ = img.shape

         y, x, _ = np.ogrid[:h, :w, :1]

         mask = np.maximum(1.0 - np.minimum(np.float32(x) / pad[0], np.float32(w - 1 - x) / pad[2]),

                           1.0 - np.minimum(np.float32(y) / pad[1], np.float32(h - 1 - y) / pad[3]))

         blur = qsize * 0.02

         img += (scipy.ndimage.gaussian_filter(img, [blur, blur, 0]) - img) * np.clip(mask * 3.0 + 1.0, 0.0, 1.0)

         img += (np.median(img, axis=(0, 1)) - img) * np.clip(mask, 0.0, 1.0)

         img = PIL.Image.fromarray(np.uint8(np.clip(np.rint(img), 0, 255)), 'RGB')

         quad += pad[:2]

     # Transform.

     img = img.transform((transform_size, transform_size), PIL.Image.QUAD, (quad + 0.5).flatten(), PIL.Image.BILINEAR)

     if output_size < transform_size:

         img = img.resize((output_size, output_size), PIL.Image.ANTIALIAS)

     # Return aligned image.

     return img