深度学习数据增强的常用方法

以下是在深度学习中经常使用的图像增强的方法

1、加噪声

from skimage.util import random_noise
    # ----1.加噪声---- #
    def _addNoise(self, img):
        '''
        输入:
            img:图像array
        输出:
            加噪声后的图像array,由于输出的像素是在[0,1]之间,所以得乘以255
        '''
        # return cv2.GaussianBlur(img, (11, 11), 0)
        return random_noise(img, mode='gaussian', clip=True) * 255

2、调整亮度

  # ---2.调整亮度--- #
    def _changeLight(self, img):
        # 从边缘分布中采样
        alpha = random.uniform(0.35, 1)
        # 做了一个零矩阵
        blank = np.zeros(img.shape, img.dtype)
        # alpha为权重，alpha的img内的像素点的值 + 1-alpha的黑颜色的值
        return cv2.addWeighted(img, alpha, blank, 1 - alpha, 0)

3、cutout

# ---3.cutout--- #
    def _cutout(self, img, bboxes, length=100, n_holes=1, threshold=0.5):
        '''
        原版本：https://github.com/uoguelph-mlrg/Cutout/blob/master/util/cutout.py
        Randomly mask out one or more patches from an image.
        Args:
            img : a 3D numpy array,(h,w,c)
            bboxes : 框的坐标
            n_holes (int): Number of patches to cut out of each image.
            length (int): The length (in pixels) of each square patch.
        '''

        def cal_iou(boxA, boxB):
            # 两张图片重叠的部分称为交集，重叠的两张图片的实际占地面积成为并集
            # IOU=交集：并集
            '''
            boxA, boxB为两个框，返回iou
            boxB为bouding box
            两张图的交集/两张图的并集
            '''
            # determine the (x, y)-coordinates of the intersection rectangle
            xA = max(boxA[0], boxB[0])
            yA = max(boxA[1], boxB[1])
            xB = min(boxA[2], boxB[2])
            yB = min(boxA[3], boxB[3])

            if xB <= xA or yB <= yA:
                return 0.0

            # compute the area of intersection rectangle
            interArea = (xB - xA + 1) * (yB - yA + 1)

            # compute the area of both the prediction and ground-truth
            # rectangles
            boxAArea = (boxA[2] - boxA[0] + 1) * (boxA[3] - boxA[1] + 1)
            boxBArea = (boxB[2] - boxB[0] + 1) * (boxB[3] - boxB[1] + 1)
            iou = interArea / float(boxBArea)
            return iou

        # 得到h和w
        if img.ndim == 3:
            h, w, c = img.shape
        else:
            _, h, w, c = img.shape
        mask = np.ones((h, w, c), np.float32)
        for n in range(n_holes):
            chongdie = True  # 看切割的区域是否与box重叠太多
            while chongdie:
                # 随机选取的x和y会决定一片区域，这片区域最后被剪掉不要了
                y = np.random.randint(h)
                x = np.random.randint(w)

                y1 = np.clip(y - length // 2, 0,
                             h)  # numpy.clip(a, a_min, a_max, out=None), clip这个函数将将数组中的元素限制在a_min, a_max之间，大于a_max的就使得它等于 a_max，小于a_min,的就使得它等于a_min
                y2 = np.clip(y + length // 2, 0, h)
                x1 = np.clip(x - length // 2, 0, w)
                x2 = np.clip(x + length // 2, 0, w)

                chongdie = False
                for box in bboxes:
                    if cal_iou([x1, y1, x2, y2], box) > threshold:
                        chongdie = True
                        break
            mask[y1: y2, x1: x2, :] = 0.
        img = img * mask
        return img

4、旋转

def flip(root_path,img_name):   #翻转图像
    img = Image.open(os.path.join(root_path, img_name))
    filp_img = img.transpose(Image.FLIP_LEFT_RIGHT)
    # filp_img.save(os.path.join(root_path,img_name.split('.')[0] + '_flip.jpg'))
    return filp_img

5、对比度增强

def contrastEnhancement(root_path, img_name):  # 对比度增强
    image = Image.open(os.path.join(root_path, img_name))
    enh_con = ImageEnhance.Contrast(image)
    # contrast = 1.1+0.4*np.random.random()#取值范围1.1-1.5
    contrast = 1.5
    image_contrasted = enh_con.enhance(contrast)
    return image_contrasted

6、仿射变化扩充图像

def fangshe_bianhuan(root_path,img_name): #仿射变化扩充图像
    img = Image.open(os.path.join(root_path, img_name))

    img = cv2.cvtColor(numpy.asarray(img) , cv2.COLOR_RGB2BGR)

    h, w = img.shape[0], img.shape[1]
    m = cv2.getRotationMatrix2D(center=(w // 2, h // 2), angle=-30, scale=0.5)
    r_img = cv2.warpAffine(src=img, M=m, dsize=(w, h), borderValue=(0, 0, 0))

    r_img = Image.fromarray(cv2.cvtColor(r_img, cv2.COLOR_BGR2RGB))
    return r_img

7、HSV数据增强

def hsv(root_path,img_name):#HSV数据增强
    h_gain , s_gain , v_gain = 0.5 , 0.5 , 0.5
    img = Image.open(os.path.join(root_path, img_name))

    img = cv2.cvtColor(numpy.asarray(img) , cv2.COLOR_RGB2BGR)

    r = np.random.uniform(-1, 1, 3) * [h_gain, s_gain, v_gain] + 1  # random gains
    hue, sat, val = cv2.split(cv2.cvtColor(img, cv2.COLOR_BGR2HSV))
    dtype = img.dtype  # uint8

    x = np.arange(0, 256, dtype=np.int16)
    lut_hue = ((x * r[0]) % 180).astype(dtype)
    lut_sat = np.clip(x * r[1], 0, 255).astype(dtype)
    lut_val = np.clip(x * r[2], 0, 255).astype(dtype)

    img_hsv = cv2.merge((cv2.LUT(hue, lut_hue), cv2.LUT(sat, lut_sat), cv2.LUT(val, lut_val))).astype(dtype)
    aug_img = cv2.cvtColor(img_hsv, cv2.COLOR_HSV2BGR)
    aug_img = Image.fromarray(cv2.cvtColor(aug_img, cv2.COLOR_BGR2RGB))
    return aug_img

8、错切变化扩充图像

def cuoqie(root_path,img_name): #错切变化扩充图像
    img = Image.open(os.path.join(root_path, img_name))

    img = cv2.cvtColor(numpy.asarray(img) , cv2.COLOR_RGB2BGR)

    h, w = img.shape[0], img.shape[1]
    origin_coord = np.array([[0, 0, 1], [w, 0, 1], [w, h, 1], [0, h, 1]])

    theta = 30  # shear角度
    tan = math.tan(math.radians(theta))

    # x方向错切
    m = np.eye(3)
    m[0, 1] = tan
    shear_coord = (m @ origin_coord.T).T.astype(np.int_)
    shear_img = cv2.warpAffine(src=img, M=m[:2],
                               dsize=(np.max(shear_coord[:, 0]), np.max(shear_coord[:, 1])),
                               borderValue=(0, 0, 0))



    c_img = Image.fromarray(cv2.cvtColor(shear_img, cv2.COLOR_BGR2RGB))
    return c_img

9、平移扩充图像，根图像移动的像素距离可自行调整，具体方法如下注释所示

def pingyi(root_path,img_name):#平移扩充图像，根图像移动的像素距离可自行调整，具体方法如下注释所示
    img = Image.open(os.path.join(root_path, img_name))
    img = cv2.cvtColor(numpy.asarray(img) , cv2.COLOR_RGB2BGR)

    cols , rows= img.shape[0], img.shape[1]
    M = np.float32([[1, 0, 50], [0, 1, 30]])#50为x即水平移动的距离，30为y 即垂直移动的距离
    dst = cv2.warpAffine(img, M, (cols, rows),borderValue=(0,255,0))
    pingyi_img = Image.fromarray(cv2.cvtColor(dst, cv2.COLOR_BGR2RGB))
    return pingyi_img

10、主函数（这里介绍如何调用前面的函数）

def createImage(imageDir,saveDir):#主函数，8种数据扩充方式，每种扩充一张
   i=0
   for name in os.listdir(imageDir):
      i=i+1
      saveName="cesun"+str(i)+".jpg"
      saveImage=contrastEnhancement(imageDir,name)
      saveImage.save(os.path.join(saveDir,saveName))
      saveName1 = "flip" + str(i) + ".jpg"
      saveImage1 = flip(imageDir,name)
      saveImage1.save(os.path.join(saveDir, saveName1))
      saveName2 = "brightnessE" + str(i) + ".jpg"
      saveImage2 = brightnessEnhancement(imageDir, name)
      saveImage2.save(os.path.join(saveDir, saveName2))
      saveName3 = "rotate" + str(i) + ".jpg"
      saveImage = rotation(imageDir, name)
      saveImage.save(os.path.join(saveDir, saveName3))
      saveName4 = "fangshe" + str(i) + ".jpg"
      saveImage = fangshe_bianhuan(imageDir, name)
      saveImage.save(os.path.join(saveDir, saveName4))
      saveName5 = "cuoqie" + str(i) + ".jpg"
      saveImage = cuoqie(imageDir, name)
      saveImage.save(os.path.join(saveDir, saveName5))
      saveName6 = "hsv" + str(i) + ".jpg"
      saveImage = hsv(imageDir, name)
      saveImage.save(os.path.join(saveDir, saveName6))
      saveName6 = "pingyi" + str(i) + ".jpg"  #不需要平移变换的，可以注释掉 这三行代码 135 136 137行
      saveImage = pingyi(imageDir, name)     #不需要平移变换的，可以注释掉 这三行代码
      saveImage.save(os.path.join(saveDir, saveName6)) #不需要平移变换的，可以注释掉 这三行代码


imageDir="jpg" #要改变的图片的路径文件夹  在当前文件夹下，建立文件夹即可
saveDir="kuochong"   #数据增强生成图片的路径文件夹
print('文件的初始文件夹为：' + imageDir)
print('----------------------------------------')
print('文件的转换后存入的文件夹为：' + saveDir)
print('----------------------------------------')
print('开始转换')
print('----------------------------------------')
createImage(imageDir,saveDir)
print('----------------------------------------')
print("数据扩充完成")