服务器之家

服务器之家 > 正文

基于Opencv制作的美颜相机带你领略美颜特效的效果

时间:2022-01-07 10:41     来源/作者:顾木子吖

 

导语

基于Opencv制作的美颜相机带你领略美颜特效的效果

​现在每一次出门,女友就喜欢拍照!BUT 嫌弃我给拍的照片角度不对,采光不好.......

基于Opencv制作的美颜相机带你领略美颜特效的效果

总之一大堆理由,啥时候让我拍照的水平能有美颜相机三分之一的效果就好!​

基于Opencv制作的美颜相机带你领略美颜特效的效果

果然都是锻炼出来的,至少现在我能看出来朋友圈哪些小姐姐批没批过照片。​

基于Opencv制作的美颜相机带你领略美颜特效的效果基于Opencv制作的美颜相机带你领略美颜特效的效果​​

​逃不掉​

基于Opencv制作的美颜相机带你领略美颜特效的效果

​逃不掉啊,为了摆脱这种局面――

立马给女友写了一款简易版本的美颜相机给她偷偷的用!这样子就不担心被锤了。机智如我.jpg

基于Opencv制作的美颜相机带你领略美颜特效的效果

 

正文

环境安装:

dlib库的安装 本博客提供三种方法进行安装  
 
T1方法:pip install dlib 此方法是需要在你安装cmake、Boost环境的计算机使用 。
T2方法:conda install -c menpo dlib=18.18此方法适合那些已经安装好conda库的环境的计算机使用。
T3方法:pip install dlib-19.8.1-cp36-cp36m-win_amd64.whl dlib库的whl文件――dlib-19.7.0-cp36-cp36m-win_amd64.rar dlib-19.3.1-cp35-cp35m-win_amd64.whl
​cv2库安装方法:
 
 pip install opencv-python

人脸五官,坐标、进行高斯模糊处理等等。

# 五官
class Organ():
	def __init__(self, img, img_hsv, temp_img, temp_hsv, landmarks, name, ksize=None):
		self.img = img
		self.img_hsv = img_hsv
		self.landmarks = landmarks
		self.name = name
		self.get_rect()
		self.shape = (int(self.bottom-self.top), int(self.right-self.left))
		self.size = self.shape[0] * self.shape[1] * 3
		self.move = int(np.sqrt(self.size/3)/20)
		self.ksize = self.get_ksize()
		self.patch_img, self.patch_hsv = self.get_patch(self.img), self.get_patch(self.img_hsv)
		self.set_temp(temp_img, temp_hsv)
		self.patch_mask = self.get_mask_relative()
	# 获取定位方框
	def get_rect(self):
		y, x = self.landmarks[:, 1], self.landmarks[:, 0]
		self.top, self.bottom, self.left, self.right = np.min(y), np.max(y), np.min(x), np.max(x)
	# 获得ksize,高斯模糊处理的参数
	def get_ksize(self, rate=15):
		size = max([int(np.sqrt(self.size/3)/rate), 1])
		size = (size if size%2==1 else size+1)
		return(size, size)
	# 截取局部切片
	def get_patch(self, img):
		shape = img.shape
		return img[np.max([self.top-self.move, 0]): np.min([self.bottom+self.move, shape[0]]), np.max([self.left-self.move, 0]): np.min([self.right+self.move, shape[1]])]
	def set_temp(self, temp_img, temp_hsv):
		self.img_temp, self.hsv_temp = temp_img, temp_hsv
		self.patch_img_temp, self.patch_hsv_temp = self.get_patch(self.img_temp), self.get_patch(self.hsv_temp)
	# 确认
	def confirm(self):
		self.img[:], self.img_hsv[:] = self.img_temp[:], self.hsv_temp[:]
	# 更新
	def update_temp(self):
		self.img_temp[:], self.hsv_temp[:] = self.img[:], self.img_hsv[:]
	# 勾画凸多边形
	def _draw_convex_hull(self, img, points, color):
		points = cv2.convexHull(points)
		cv2.fillConvexPoly(img, points, color=color)
	# 获得局部相对坐标遮盖
	def get_mask_relative(self, ksize=None):
		if ksize == None:
			ksize = self.ksize
		landmarks_re = self.landmarks.copy()
		landmarks_re[:, 1] -= np.max([self.top-self.move, 0])
		landmarks_re[:, 0] -= np.max([self.left-self.move, 0])
		mask = np.zeros(self.patch_img.shape[:2], dtype=np.float64)
		self._draw_convex_hull(mask, landmarks_re, color=1)
		mask = np.array([mask, mask, mask]).transpose((1, 2, 0))
		mask = (cv2.GaussianBlur(mask, ksize, 0) > 0) * 1.0
		return cv2.GaussianBlur(mask, ksize, 0)[:]
	# 获得全局绝对坐标遮盖
	def get_mask_abs(self, ksize=None):
		if ksize == None:
			ksize = self.ksize
		mask = np.zeros(self.img.shape, dtype=np.float64)
		patch = self.get_patch(mask)
		patch[:] = self.patch_mask[:]
		return mask

基于Opencv制作的美颜相机带你领略美颜特效的效果主要美颜效果进行的处理如下:

# 美白
	def whitening(self, rate=0.15, confirm=True):
		if confirm:
			self.confirm()
			self.patch_hsv[:, :, -1] = np.minimum(self.patch_hsv[:, :, -1]+self.patch_hsv[:, :, -1]*self.patch_mask[:, :, -1]*rate, 255).astype("uint8")
			self.img[:]=cv2.cvtColor(self.img_hsv, cv2.COLOR_HSV2BGR)[:]
			self.update_temp()
		else:
			self.patch_hsv_temp[:] = cv2.cvtColor(self.patch_img_temp, cv2.COLOR_BGR2HSV)[:]
			self.patch_hsv_temp[:, :, -1] = np.minimum(self.patch_hsv_temp[:, :, -1]+self.patch_hsv_temp[:, :, -1]*self.patch_mask[:, :, -1]*rate, 255).astype("uint8")
			self.patch_img_temp[:] = cv2.cvtColor(self.patch_hsv_temp, cv2.COLOR_HSV2BGR)[:]
	# 提升鲜艳度
	def brightening(self, rate=0.3, confirm=True):
		patch_mask = self.get_mask_relative((1, 1))
		if confirm:
			self.confirm()
			patch_new = self.patch_hsv[:, :, 1]*patch_mask[:, :, 1]*rate
			patch_new = cv2.GaussianBlur(patch_new, (3, 3), 0)
			self.patch_hsv[:, :, 1] = np.minimum(self.patch_hsv[:, :, 1]+patch_new, 255).astype("uint8")
			self.img[:]=cv2.cvtColor(self.img_hsv, cv2.COLOR_HSV2BGR)[:]
			self.update_temp()
		else:
			self.patch_hsv_temp[:] = cv2.cvtColor(self.patch_img_temp, cv2.COLOR_BGR2HSV)[:]
			patch_new = self.patch_hsv_temp[:, :, 1]*patch_mask[:, :, 1]*rate
			patch_new = cv2.GaussianBlur(patch_new, (3, 3), 0)
			self.patch_hsv_temp[:, :, 1] = np.minimum(self.patch_hsv[:, :, 1]+patch_new, 255).astype("uint8")
			self.patch_img_temp[:] = cv2.cvtColor(self.patch_hsv_temp, cv2.COLOR_HSV2BGR)[:]
	# 磨平
	def smooth(self, rate=0.6, ksize=None, confirm=True):
		if ksize == None:
			ksize=self.get_ksize(80)
		index = self.patch_mask > 0
		if confirm:
			self.confirm()
			patch_new = cv2.GaussianBlur(cv2.bilateralFilter(self.patch_img, 3, *ksize), ksize, 0)
			self.patch_img[index] = np.minimum(rate*patch_new[index]+(1-rate)*self.patch_img[index], 255).astype("uint8")
			self.img_hsv[:] = cv2.cvtColor(self.img, cv2.COLOR_BGR2HSV)[:]
			self.update_temp()
		else:
			patch_new = cv2.GaussianBlur(cv2.bilateralFilter(self.patch_img_temp, 3, *ksize), ksize, 0)
			self.patch_img_temp[index] = np.minimum(rate*patch_new[index]+(1-rate)*self.patch_img_temp[index], 255).astype("uint8")
			self.patch_hsv_temp[:] = cv2.cvtColor(self.patch_img_temp, cv2.COLOR_BGR2HSV)[:]
	# 锐化
	def sharpen(self, rate=0.3, confirm=True):
		patch_mask = self.get_mask_relative((3, 3))
		kernel = np.zeros((9, 9), np.float32)
		kernel[4, 4] = 2.0
		boxFilter = np.ones((9, 9), np.float32) / 81.0
		kernel = kernel - boxFilter
		index = patch_mask > 0
		if confirm:
			self.confirm()
			sharp = cv2.filter2D(self.patch_img, -1, kernel)
			self.patch_img[index] = np.minimum(((1-rate)*self.patch_img)[index]+sharp[index]*rate, 255).astype("uint8")
			self.update_temp()
		else:
			sharp = cv2.filter2D(self.patch_img_temp, -1, kernel)
			self.patch_img_temp[:] = np.minimum(self.patch_img_temp+self.patch_mask*sharp*rate, 255).astype("uint8")
			self.patch_hsv_temp[:] = cv2.cvtColor(self.patch_img_temp, cv2.COLOR_BGR2HSV)[:]
            
 
# 额头
class ForeHead(Organ):
	def __init__(self, img, img_hsv, temp_img, temp_hsv, landmarks, mask_organs, name, ksize=None):
		self.mask_organs = mask_organs
		super(ForeHead, self).__init__(img, img_hsv, temp_img, temp_hsv, landmarks, name, ksize)
	# 获得局部相对坐标mask
	def get_mask_relative(self, ksize=None):
		if ksize == None:
			ksize = self.ksize
		landmarks_re = self.landmarks.copy()
		landmarks_re[:, 1] -= np.max([self.top-self.move, 0])
		landmarks_re[:, 0] -= np.max([self.left-self.move, 0])
		mask = np.zeros(self.patch_img.shape[:2], dtype=np.float64)
		self._draw_convex_hull(mask, landmarks_re, color=1)
		mask = np.array([mask, mask, mask]).transpose((1, 2, 0))
		mask = (cv2.GaussianBlur(mask, ksize, 0) > 0) * 1.0
		patch_organs = self.get_patch(self.mask_organs)
		mask= cv2.GaussianBlur(mask, ksize, 0)[:]
		mask[patch_organs>0] = (1-patch_organs[patch_organs>0])
		return mask
 
 
# 脸类
class Face(Organ):
	def __init__(self, img, img_hsv, temp_img, temp_hsv, landmarks, index):
		self.index = index
		# 五官:下巴、嘴、鼻子、左右眼、左右耳
		self.organs_name = ["jaw", "mouth", "nose", "left_eye", "right_eye", "left_brow", "right_brow"]
		# 五官标记点
		self.organs_point = [list(range(0, 17)), list(range(48, 61)), 
							 list(range(27, 35)), list(range(42, 48)), 
							 list(range(36, 42)), list(range(22, 27)),
							 list(range(17, 22))]
		self.organs = {name: Organ(img, img_hsv, temp_img, temp_hsv, landmarks[points], name) for name, points in zip(self.organs_name, self.organs_point)}
		# 额头
		mask_nose = self.organs["nose"].get_mask_abs()
		mask_organs = (self.organs["mouth"].get_mask_abs()+mask_nose+self.organs["left_eye"].get_mask_abs()+self.organs["right_eye"].get_mask_abs()+self.organs["left_brow"].get_mask_abs()+self.organs["right_brow"].get_mask_abs())
		forehead_landmark = self.get_forehead_landmark(img, landmarks, mask_organs, mask_nose)
		self.organs["forehead"] = ForeHead(img, img_hsv, temp_img, temp_hsv, forehead_landmark, mask_organs, "forehead")
		mask_organs += self.organs["forehead"].get_mask_abs()
		# 人脸的完整标记点
		self.FACE_POINTS = np.concatenate([landmarks, forehead_landmark])
		super(Face, self).__init__(img, img_hsv, temp_img, temp_hsv, self.FACE_POINTS, "face")
		mask_face = self.get_mask_abs() - mask_organs
		self.patch_mask = self.get_patch(mask_face)
	# 计算额头坐标
	def get_forehead_landmark(self, img, face_landmark, mask_organs, mask_nose):
		radius = (np.linalg.norm(face_landmark[0]-face_landmark[16])/2).astype("int32")
		center_abs = tuple(((face_landmark[0]+face_landmark[16])/2).astype("int32"))
		angle = np.degrees(np.arctan((lambda l:l[1]/l[0])(face_landmark[16]-face_landmark[0]))).astype("int32")
		mask = np.zeros(mask_organs.shape[:2], dtype=np.float64)
		cv2.ellipse(mask, center_abs, (radius, radius), angle, 180, 360, 1, -1)
		# 剔除与五官重合部分
		mask[mask_organs[:, :, 0]>0]=0
		# 根据鼻子的肤色判断真正的额头面积
		index_bool = []
		for ch in range(3):
			mean, std = np.mean(img[:, :, ch][mask_nose[:, :, ch]>0]), np.std(img[:, :, ch][mask_nose[:, :, ch]>0])
			up, down = mean+0.5*std, mean-0.5*std
			index_bool.append((img[:, :, ch]<down)|(img[:, :, ch]>up))
		index_zero = ((mask>0)&index_bool[0]&index_bool[1]&index_bool[2])
		mask[index_zero] = 0
		index_abs = np.array(np.where(mask>0)[::-1]).transpose()
		landmark = cv2.convexHull(index_abs).squeeze()
		return landmark
 
 
# 化妆器
class Makeup():
	def __init__(self, predictor_path="./predictor/shape_predictor_68_face_landmarks.dat"):
		self.photo_path = []
		self.predictor_path = predictor_path
		self.faces = {}
		# 人脸检测与特征提取
		self.detector = dlib.get_frontal_face_detector()
		self.predictor = dlib.shape_predictor(self.predictor_path)
	# 人脸定位和特征提取
	# img为numpy数组
	# 返回值为人脸特征(x, y)坐标的矩阵
	def get_faces(self, img, img_hsv, temp_img, temp_hsv, name, n=1):
		rects = self.detector(img, 1)
		if len(rects) < 1:
			print("[Warning]:No face detected...")
			return None
		return {name: [Face(img, img_hsv, temp_img, temp_hsv, np.array([[p.x, p.y] for p in self.predictor(img, rect).parts()]), i) for i, rect in enumerate(rects)]}
	# 读取图片
	def read_img(self, fname, scale=1):
		img = cv2.imdecode(np.fromfile(fname, dtype=np.uint8), -1)
		if not type(img):
			print("[ERROR]:Fail to Read %s" % fname)
			return None
		return img
	def read_and_mark(self, fname):
		img = self.read_img(fname)
		img_hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
		temp_img, temp_hsv = img.copy(), img_hsv.copy()
		return img, temp_img, self.get_faces(img, img_hsv, temp_img, temp_hsv, fname)

效果如下:

基于Opencv制作的美颜相机带你领略美颜特效的效果

基于Opencv制作的美颜相机带你领略美颜特效的效果

基于Opencv制作的美颜相机带你领略美颜特效的效果​​嘿嘿――小姐姐美颜之后是不是白了很多吖!

 

总结

本次文章就到这里啦!如需完整的打包好的项目源码基地见:#私信小编06#即可免费领取!

记得关注、评论、点赞三连哦~

基于Opencv制作的美颜相机带你领略美颜特效的效果

到此这篇关于基于Opencv制作的美颜相机带你领略美颜特效的效果的文章就介绍到这了,更多相关Opencv 美颜相机内容请搜索服务器之家以前的文章或继续浏览下面的相关文章希望大家以后多多支持服务器之家!

原文链接:https://blog.csdn.net/weixin_55822277/article/details/120333550

相关文章

热门资讯

蜘蛛侠3英雄无归3正片免费播放 蜘蛛侠3在线观看免费高清完整
蜘蛛侠3英雄无归3正片免费播放 蜘蛛侠3在线观看免费高清完整 2021-08-24
yue是什么意思 网络流行语yue了是什么梗
yue是什么意思 网络流行语yue了是什么梗 2020-10-11
背刺什么意思 网络词语背刺是什么梗
背刺什么意思 网络词语背刺是什么梗 2020-05-22
2020微信伤感网名听哭了 让对方看到心疼的伤感网名大全
2020微信伤感网名听哭了 让对方看到心疼的伤感网名大全 2019-12-26
2021年耽改剧名单 2021要播出的59部耽改剧列表
2021年耽改剧名单 2021要播出的59部耽改剧列表 2021-03-05
返回顶部