(2024年11月12日更新）图片清晰度批量分级筛选工具--by iyuandou

lhs · 发表于 2025-1-22 21:22:34

建议导入DFLIMG类进行人脸区域确定。
使用          face_mask = LandmarksProcessor.get_image_hull_mask(image.shape, landmarks)获得遮罩，通过这个遮罩计算 x, y, w, h 到face_img。
在对face_img进行方差计算，避免产生意外的方差值，和排除面部外的区域。如下所示：
import cv2
from pathlib import Path
from facelib import LandmarksProcessor
from DFLIMG import DFLIMG

class FaceMaskDisplay:
def __init__(self, input_dir):
      self.input_dir = input_dir
      self.image_paths = self.load_image_paths()
      self.images = []
      self.dflimgs = []
      self.landmarks_list = []
      self.face_masks = []
      self.variances = []

def load_image_paths(self):
      input_dir = Path(self.input_dir)
      image_paths = list(input_dir.glob("*.jpg")) + list(input_dir.glob("*.png"))
      return image_paths

def load_images_and_dflimgs(self):
      for image_path in self.image_paths:
         image = cv2.imread(str(image_path))
         self.images.append(image)
         dflimg = DFLIMG.load(image_path)
         if dflimg is None or not dflimg.has_data():
            print(f"{image_path} 不符合DFL规范")
            self.dflimgs.append(None)
         else:
            self.dflimgs.append(dflimg)

def get_facial_areas(self):
      for i, dflimg in enumerate(self.dflimgs):
         if dflimg is None:
            self.landmarks_list.append(None)
            self.face_masks.append(None)
            continue
         landmarks = dflimg.get_landmarks()
         self.landmarks_list.append(landmarks)
         image = self.images[i]
         face_mask = LandmarksProcessor.get_image_hull_mask(image.shape, landmarks)
         self.face_masks.append(face_mask)

def variance_of_laplacian(self, image):
      """计算图像的拉普拉斯方差"""
      gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
      return cv2.Laplacian(gray, cv2.CV_64F).var()

def compute_variances_and_display(self):
      for i, (image_path, face_mask) in enumerate(zip(self.image_paths, self.face_masks)):
         if face_mask is None:
            self.variances.append(None)
            continue
         # 计算人脸区域的矩形框坐标 (x, y, w, h)
         x, y, w, h = cv2.boundingRect(self.landmarks_list[i].astype(int))
         # 提取人脸区域内容
         face_img = self.images[i][y:y+h, x:x+w]
         variance = self.variance_of_laplacian(face_img)
         self.variances.append(variance)
         # 显示图像的面部区域
         cv2.imshow('Face Image', face_img)
         print(f"图像 {image_path.name} 的拉普拉斯方差为: {variance}")
         # 等待用户关闭窗口
         key = cv2.waitKey(0)
         cv2.destroyAllWindows()
         # 当用户按下任意键后进行分类和保存
         if key != -1:
            self.classify_and_save_image(image_path, variance)

def classify_and_save_image(self, image_path, variance):
      idx = int(variance / 10) * 10  # 根据方差值确定输出目录索引
      if idx >= 100:
         idx = 100
      output_dir = Path(self.input_dir) / f"variance_{idx}-{idx+9}"
      output_dir.mkdir(parents=True, exist_ok=True)
      output_path = output_dir / image_path.name
      cv2.imwrite(str(output_path), cv2.imread(str(image_path)))

# 示例用法
if __name__ == "__main__":
input_dir = r"C:\Users\adnain\Desktop\temp3"
face_mask_display = FaceMaskDisplay(input_dir)
face_mask_display.load_images_and_dflimgs()
face_mask_display.get_facial_areas()
face_mask_display.compute_variances_and_display()