|
|
|
|
|
import cv2
|
|
|
|
|
|
import numpy as np
|
|
|
|
|
|
import albumentations as A
|
|
|
|
|
|
|
|
|
|
|
|
#图像处理类
|
|
|
|
|
|
class Image_enhancement:
|
|
|
|
|
|
def __init__(self,):
|
|
|
|
|
|
pass
|
|
|
|
|
|
"""===============================1.几何处理:=================================="""
|
|
|
|
|
|
# 1.1 图片旋转(可选择旋转角度)
|
|
|
|
|
|
def roate(self, frame=None,angle=30):
|
|
|
|
|
|
h, w = frame.shape[:2]
|
|
|
|
|
|
M = cv2.getRotationMatrix2D((w / 2, h / 2), angle, 1)
|
|
|
|
|
|
img_roate = cv2.warpAffine(frame, M, (w, h))
|
|
|
|
|
|
return img_roate
|
|
|
|
|
|
|
|
|
|
|
|
# 1.2 图片缩放(可选择缩放大小)
|
|
|
|
|
|
def resize(self, frame=None,w=224,h=224):
|
|
|
|
|
|
img_resized = cv2.resize(frame, (w, h)) # w,h
|
|
|
|
|
|
return img_resized
|
|
|
|
|
|
|
|
|
|
|
|
"""===============================2.颜色空间变化:=================================="""
|
|
|
|
|
|
# 2.1 图片的对比度、饱和度、亮度、色调变化
|
|
|
|
|
|
def ColorJitter(self, frame=None,brightness=0,hue=0,saturation=0,contrast=0):
|
|
|
|
|
|
transform = A.ColorJitter(brightness=float(brightness),hue=float(hue),saturation=float(saturation),contrast=float(contrast),p=1)
|
|
|
|
|
|
img_result = transform(image=np.array(frame))
|
|
|
|
|
|
img_color = img_result['image']
|
|
|
|
|
|
return img_color
|
|
|
|
|
|
|
|
|
|
|
|
# 2.2 灰度图转换
|
|
|
|
|
|
def ToGray(self, frame=None ):
|
|
|
|
|
|
# cvtColor的第一个参数是处理的图像，第二个是RGB2GRAY
|
|
|
|
|
|
gray_img = cv2.cvtColor(frame, cv2.COLOR_RGB2GRAY)
|
|
|
|
|
|
return gray_img
|
|
|
|
|
|
|
|
|
|
|
|
# 2.3 直方图均衡化
|
|
|
|
|
|
def equalhist(self, frame=None):
|
|
|
|
|
|
# 直方图均衡化增强对比度:通过调整图像的灰度分布，增加图像的对比度，从而使图像更加清晰。
|
|
|
|
|
|
frame_lab = cv2.cvtColor(frame, cv2.COLOR_BGR2LAB)
|
|
|
|
|
|
frame_lab[:, :, 0] = cv2.equalizeHist(frame_lab[:, :, 0])
|
|
|
|
|
|
enhanced_frame = cv2.cvtColor(frame_lab, cv2.COLOR_LAB2BGR)
|
|
|
|
|
|
return enhanced_frame
|
|
|
|
|
|
|
|
|
|
|
|
"""===============================3.频率像素点操作:=================================="""
|
|
|
|
|
|
# 3.1 模糊
|
|
|
|
|
|
# 3.1.1 高斯滤波
|
|
|
|
|
|
def Gaussblur(self, frame=None,count=3):
|
|
|
|
|
|
transform = A.GaussianBlur(blur_limit=count, p=1)
|
|
|
|
|
|
img_result = transform(image=np.array(frame))['image']
|
|
|
|
|
|
return img_result
|
|
|
|
|
|
|
|
|
|
|
|
# 3.1.3 随机模糊
|
|
|
|
|
|
def Blur(self, frame=None, count=3):
|
|
|
|
|
|
transform = A.Blur(blur_limit=count, p=1)
|
|
|
|
|
|
img_result = transform(image=np.array(frame))['image']
|
|
|
|
|
|
return img_result
|
|
|
|
|
|
|
|
|
|
|
|
# 3.1.2 中值滤波
|
|
|
|
|
|
def Medianblur(self, frame=None, count=3):
|
|
|
|
|
|
transform = A.MedianBlur(blur_limit=count, p=1)
|
|
|
|
|
|
img_result = transform(image=np.array(frame))['image']
|
|
|
|
|
|
return img_result
|
|
|
|
|
|
|
|
|
|
|
|
# 3.2 锐化
|
|
|
|
|
|
# 3.2.1 使用sobel算子进行锐化
|
|
|
|
|
|
def sobel(self, frame=None):
|
|
|
|
|
|
img_gray = cv2.cvtColor(frame, cv2.COLOR_RGB2GRAY)
|
|
|
|
|
|
kernelx1 = np.array([[-1, -2, -1], [0, 0, 0], [1, 2, 1]], dtype=int)
|
|
|
|
|
|
kernely1 = np.array([[-1, 0, 1], [-2, 0, 2], [-1, 0, 1]], dtype=int)
|
|
|
|
|
|
x1 = cv2.filter2D(img_gray, cv2.CV_16S, kernelx1)
|
|
|
|
|
|
y1 = cv2.filter2D(img_gray, cv2.CV_16S, kernely1)
|
|
|
|
|
|
absX1 = cv2.convertScaleAbs(x1)
|
|
|
|
|
|
absY1 = cv2.convertScaleAbs(y1)
|
|
|
|
|
|
img_result = cv2.addWeighted(absX1, 0.5, absY1, 0.5, 0)
|
|
|
|
|
|
return img_result
|
|
|
|
|
|
|
|
|
|
|
|
# 3.2.2 使用Prewitt算子进行锐化
|
|
|
|
|
|
def Prewitt(self, frame=None):
|
|
|
|
|
|
img_gray = cv2.cvtColor(frame, cv2.COLOR_RGB2GRAY)
|
|
|
|
|
|
kernelx1 = np.array([[-1, 0, 1], [-1, 0, 1], [-1, 0, 1]], dtype=int)
|
|
|
|
|
|
kernely1 = np.array([[-1, -1, 1], [0, 0, 0], [1, 1, 1]], dtype=int)
|
|
|
|
|
|
x1 = cv2.filter2D(img_gray, cv2.CV_16S, kernelx1)
|
|
|
|
|
|
y1 = cv2.filter2D(img_gray, cv2.CV_16S, kernely1)
|
|
|
|
|
|
absX1 = cv2.convertScaleAbs(x1)
|
|
|
|
|
|
absY1 = cv2.convertScaleAbs(y1)
|
|
|
|
|
|
img_result = cv2.addWeighted(absX1, 0.5, absY1, 0.5, 0)
|
|
|
|
|
|
return img_result
|
|
|
|
|
|
|
|
|
|
|
|
# 3.2.3 使用robert算子进行锐化
|
|
|
|
|
|
def robert(self, frame=None):
|
|
|
|
|
|
img_gray = cv2.cvtColor(frame, cv2.COLOR_RGB2GRAY)
|
|
|
|
|
|
kernelx1 = np.array([[-1, 0], [0, 1]], dtype=int)
|
|
|
|
|
|
kernely1 = np.array([[0, -1], [1, 0]], dtype=int)
|
|
|
|
|
|
x1 = cv2.filter2D(img_gray, cv2.CV_16S, kernelx1)
|
|
|
|
|
|
y1 = cv2.filter2D(img_gray, cv2.CV_16S, kernely1)
|
|
|
|
|
|
absX1 = cv2.convertScaleAbs(x1)
|
|
|
|
|
|
absY1 = cv2.convertScaleAbs(y1)
|
|
|
|
|
|
img_result = cv2.addWeighted(absX1, 0.5, absY1, 0.5, 0)
|
|
|
|
|
|
return img_result
|
|
|
|
|
|
|
|
|
|
|
|
# 3.3 添加噪声
|
|
|
|
|
|
# 3.3.1 添加高斯噪声
|
|
|
|
|
|
def add_gaussian_noise(self, frame=None, mean=0,sigma=30):
|
|
|
|
|
|
# 生成高斯噪声矩阵
|
|
|
|
|
|
row, col, ch = frame.shape
|
|
|
|
|
|
gaussian = np.random.randn(row, col, ch) * sigma + mean
|
|
|
|
|
|
gaussian = gaussian.reshape(row, col, ch)
|
|
|
|
|
|
img_result = frame + gaussian
|
|
|
|
|
|
# 转换数据类型为8位无符号整数类型
|
|
|
|
|
|
img_result = cv2.convertScaleAbs(img_result)
|
|
|
|
|
|
return img_result
|
|
|
|
|
|
# 3.3.2 添加椒盐噪声
|
|
|
|
|
|
def add_salt_and_pepper_noise(self, frame=None, percentage=0): # percentage [0,100]
|
|
|
|
|
|
# 确保百分比在 0 到 100 之间
|
|
|
|
|
|
if percentage < 0 or percentage > 100:
|
|
|
|
|
|
percentage=0
|
|
|
|
|
|
|
|
|
|
|
|
# 生成椒盐噪声矩阵
|
|
|
|
|
|
row, col, ch = frame.shape
|
|
|
|
|
|
noise = np.zeros((row, col, ch), np.uint8)
|
|
|
|
|
|
for i in range(row):
|
|
|
|
|
|
for j in range(col):
|
|
|
|
|
|
rand = np.random.randint(0, 100)
|
|
|
|
|
|
if rand < percentage:
|
|
|
|
|
|
noise[i][j] = [0, 0, 0]
|
|
|
|
|
|
elif rand > 100 - percentage:
|
|
|
|
|
|
noise[i][j] = [255, 255, 255]
|
|
|
|
|
|
else:
|
|
|
|
|
|
noise[i][j] = frame[i][j]
|
|
|
|
|
|
|
|
|
|
|
|
# 将椒盐噪声矩阵添加到原始图像中
|
|
|
|
|
|
img_result = cv2.add(frame, noise)
|
|
|
|
|
|
return img_result
|
|
|
|
|
|
# 3.3.3 添加均值噪声
|
|
|
|
|
|
def add_mean_noise(self, frame=None, mean=0,std_dev=30):
|
|
|
|
|
|
# 生成均值噪声矩阵
|
|
|
|
|
|
row, col, ch = frame.shape
|
|
|
|
|
|
noise = np.random.normal(mean, std_dev, (row, col, ch)).astype(np.uint8)
|
|
|
|
|
|
# 将均值噪声矩阵添加到原始图像中
|
|
|
|
|
|
img_result = cv2.add(frame, noise)
|
|
|
|
|
|
return img_result
|
|
|
|
|
|
|
|
|
|
|
|
# 3.4 边缘检测
|
|
|
|
|
|
def Canny(self, frame=None):
|
|
|
|
|
|
# 转为灰度图
|
|
|
|
|
|
gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
|
|
|
|
|
|
# 高斯滤波
|
|
|
|
|
|
gray = cv2.GaussianBlur(gray, (5, 5), 0)
|
|
|
|
|
|
# 边缘检测
|
|
|
|
|
|
canny_edged = cv2.Canny(gray, 30, 250)
|
|
|
|
|
|
return canny_edged
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
if __name__ == "__main__":
|
|
|
|
|
|
# 类实例化
|
|
|
|
|
|
img_enhance = Image_enhancement()
|
|
|
|
|
|
|
|
|
|
|
|
img_path = r'images/car_test.jpg'
|
|
|
|
|
|
img = cv2.imread(img_path)
|
|
|
|
|
|
|
|
|
|
|
|
# 测试1.1 图片旋转(可选择旋转角度)
|
|
|
|
|
|
img_roate = img_enhance.roate(img,45)
|
|
|
|
|
|
cv2.imwrite("out_imgs/out_1.jpg",img_roate)
|
|
|
|
|
|
# 测试1.2 图片缩放(可选择缩放大小)
|
|
|
|
|
|
img_resized = img_enhance.resize(img, w=224, h=224)
|
|
|
|
|
|
cv2.imwrite("out_imgs/out_2.jpg", img_resized)
|
|
|
|
|
|
|
|
|
|
|
|
# 测试2.1 图片的对比度、饱和度、亮度、色调变化
|
|
|
|
|
|
img_resized = img_enhance.ColorJitter(img, brightness=10,hue=10,saturation=10,contrast=10)
|
|
|
|
|
|
cv2.imwrite("out_imgs/out_3.jpg", img_resized)
|
|
|
|
|
|
# 测试2.2 灰度图转化
|
|
|
|
|
|
gray_img = img_enhance.ToGray(img)
|
|
|
|
|
|
cv2.imwrite("out_imgs/out_4.jpg", gray_img)
|
|
|
|
|
|
# 测试2.3 直方图均衡化
|
|
|
|
|
|
enhanced_frame = img_enhance.equalhist(img)
|
|
|
|
|
|
cv2.imwrite("out_imgs/out_5.jpg", enhanced_frame)
|
|
|
|
|
|
|
|
|
|
|
|
# 3.1 模糊
|
|
|
|
|
|
# 3.1.1 高斯滤波
|
|
|
|
|
|
img_gauss = img_enhance.Gaussblur(img,count=5)
|
|
|
|
|
|
cv2.imwrite("out_imgs/out_6.jpg", img_gauss)
|
|
|
|
|
|
|
|
|
|
|
|
# 3.1.2 随机模糊
|
|
|
|
|
|
img_Blur = img_enhance.Blur(img, count=5)
|
|
|
|
|
|
cv2.imwrite("out_imgs/out_7.jpg", img_Blur)
|
|
|
|
|
|
|
|
|
|
|
|
# 3.1.3 中值滤波
|
|
|
|
|
|
img_Medianblur = img_enhance.Medianblur(img, count=5)
|
|
|
|
|
|
cv2.imwrite("out_imgs/out_8.jpg", img_Medianblur)
|
|
|
|
|
|
|
|
|
|
|
|
# 3.2 锐化
|
|
|
|
|
|
# 3.2.1 使用sobel算子进行锐化
|
|
|
|
|
|
img_sobel = img_enhance.sobel(img)
|
|
|
|
|
|
cv2.imwrite("out_imgs/out_9.jpg", img_sobel)
|
|
|
|
|
|
# 3.2.2 使用Prewitt算子进行锐化
|
|
|
|
|
|
img_Prewitt = img_enhance.Prewitt(img)
|
|
|
|
|
|
cv2.imwrite("out_imgs/out_10.jpg", img_Prewitt)
|
|
|
|
|
|
# 3.2.3 使用robert算子进行锐化
|
|
|
|
|
|
img_robert = img_enhance.robert(img)
|
|
|
|
|
|
cv2.imwrite("out_imgs/out_11.jpg", img_robert)
|
|
|
|
|
|
|
|
|
|
|
|
# 3.3 添加噪声
|
|
|
|
|
|
# 3.3.1 添加高斯噪声
|
|
|
|
|
|
img_gaussian_noise = img_enhance.add_gaussian_noise(img,mean=0,sigma=30)
|
|
|
|
|
|
cv2.imwrite("out_imgs/out_12.jpg", img_gaussian_noise)
|
|
|
|
|
|
# 3.3.2 添加椒盐噪声
|
|
|
|
|
|
img_salt_and_pepper_noise = img_enhance.add_salt_and_pepper_noise(img, percentage=10)
|
|
|
|
|
|
cv2.imwrite("out_imgs/out_13.jpg", img_salt_and_pepper_noise)
|
|
|
|
|
|
# 3.3.3 添加均值噪声
|
|
|
|
|
|
img_mean_noise = img_enhance.add_mean_noise(img, mean=0, std_dev=30)
|
|
|
|
|
|
cv2.imwrite("out_imgs/out_14.jpg", img_mean_noise)
|
|
|
|
|
|
|
|
|
|
|
|
# 3.4 边缘检测
|
|
|
|
|
|
img_Canny = img_enhance.Canny(img)
|
|
|
|
|
|
cv2.imwrite("out_imgs/out_15.jpg", img_Canny)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
