使用Python与OpenCV为图像增加高斯噪声，雾霾等降质因素 September 11, 2018 使用Python与OpenCV为图像增加高斯噪声，雾霾，低照度，高斯模糊，JPG压缩失真降质因素。 环境要求：Python3.5+, opencv-python3.4+ <!--more--> ```python # coding = utf-8 # auth: pizi # email: wangke_0809@126.com import numpy as np import cv2 def add_gasuss_noise(image, mean=0, var=0.001): ''' 添加高斯噪声 mean : 均值 var : 方差 ''' image = np.array(image/255, dtype=float) noise = np.random.normal(mean, var ** 0.5, image.shape) out = image + noise if out.min() < 0: low_clip = -1. else: low_clip = 0. out = np.clip(out, low_clip, 1.0) out = np.uint8(out*255) return out def add_haze(image, t=0.6, A=1): ''' 添加雾霾 t : 透视率 0~1 A : 大气光照 ''' out = image*t + A*255*(1-t) return out def ajust_image(image, cont=1, bright=0): ''' 调整对比度与亮度 cont : 对比度，调节对比度应该与亮度同时调节 bright : 亮度 ''' out = np.uint8(np.clip((cont * image + bright), 0, 255)) # tmp = np.hstack((img, res)) # 两张图片横向合并（便于对比显示） return out def ajust_image_hsv(image, h=1, s=1, v=0.8): ''' 调整HSV通道，调整V通道以调整亮度 各通道系数 ''' HSV = cv2.cvtColor(image, cv2.COLOR_BGR2HSV) H, S, V = cv2.split(HSV) H2 = np.uint8(H * h) S2 = np.uint8(S * s) V2 = np.uint8(V * v) hsv_image = cv2.merge([H2, S2, V2]) out = cv2.cvtColor(hsv_image, cv2.COLOR_HSV2BGR) return out def ajust_jpg_quality(image, q=100, save_path=None): ''' 调整图像JPG压缩失真程度 q : 压缩质量 0~100 ''' if save_path is None: cv2.imwrite("jpg_tmp.jpg", image, [int(cv2.IMWRITE_JPEG_QUALITY), q]) out = cv2.imread('jpg_tmp.jpg') return out else: cv2.imwrite(save_path, image, [int(cv2.IMWRITE_JPEG_QUALITY), q]) def add_gasuss_blur(image, kernel_size=(3, 3), sigma=0.1): ''' 添加高斯模糊 kernel_size : 模糊核大小 sigma : 标准差 ''' out = cv2.GaussianBlur(image, kernel_size, sigma) return out def test_methods(): img = cv2.imread('test.jpg') out = add_haze(img) cv2.imwrite("add_haze.jpg", out) out = add_gasuss_noise(img) cv2.imwrite("add_gasuss_noise.jpg", out) out = add_gasuss_blur(img) cv2.imwrite("add_gasuss_blur.jpg", out) out = ajust_image(img) cv2.imwrite("ajust_image.jpg", out) out = ajust_image_hsv(img) cv2.imwrite("ajust_image_hsv.jpg", out) ajust_jpg_quality(img, save_path='ajust_jpg_quality.jpg') test_methods() ``` 针对雾霾，如果已知深度图，如 NYU Depth V2 等数据集，利用单散射模型增加雾霾。