【算数变换】椒盐噪声(Salt Pepper Noise)

原理

1. 概念：在图像上随机地添加了黑色和白色地像素点，使得图像看起来像是撒了椒盐。
   • 盐噪声：白色噪声，是高灰度噪声
   • 椒噪声：黑色噪声，是低灰度噪声
2. 去除方法：中值滤波

实战

def imgaug_example(image, p=0.03):
    """
    args:
        image: np.ndarray
        p: 将像素替换为椒盐噪声的概率。 取值范围(0.0, 0.03)
    """
    augmenter = iaa.SaltAndPepper(p=p)
    image = augmenter(image=image)
    return image


def torch_example(image, p=0.03):
    """
    args:
        image: torch.Tensor
        p: 将像素替换为椒盐噪声的概率。 取值范围(0.0, 0.03)
    """

    _, imh, imw = image.shape
    num_salt = int(p * imh * imw)

    num_salt_0 = random.randint(0, num_salt)
    num_salt_255 = num_salt - num_salt_0

    coords1 = [torch.randint(low=0, high=i - 1, size=(num_salt_0,)) for i in [imh, imw]]
    image[:, coords1[0], coords1[1]] = 0

    coords2 = [torch.randint(low=0, high=i - 1, size=(num_salt_255,)) for i in [imh, imw]]
    image[:, coords2[0], coords2[1]] = 255

    return image