面向深度学习的图像数据增强综述A survey on Image Data Augmentation for Deep Learning

    import tensorlayer as tl
    image = tl.vis.read_image(r"2.jpg")
    
    #缩放，缩小0.8倍  
    xx = tl.prepro.zoom(image, zoom_range=0.8)
    #axis=1,水平翻转，axis=0,垂直翻转
    xx = tl.prepro.flip_axis(image, axis=1, is_random=False)
    # 调整亮度 gamma小于1，变亮，越小越亮
    xx =tl.prepro.brightness(image, gamma=1.2, is_random=False)
    #裁剪
    xx=tl.prepro.crop(image, 400,400, is_random=False, row_index=0, col_index=1)
    #调整亮度，饱和度，对比度
     xx=tl.prepro.illumination(image, gamma=0.7, contrast=1.0 , saturation=1, is_random=False)
    #随机调整亮度，饱和度，对比度
    xx = tl.prepro.illumination(image, gamma=(0.5, 5.0), contrast=(0.3, 1.0), saturation=(0.7, 1.0), is_random=True)
    #平移
    xx = tl.prepro.shift(xx, wrg=-0.1, hrg=0, is_random=False)
    #噪声
    xx=tl.prepro.drop(image, keep=0.5)
    #色彩空间转化rgb->hsv
    xx=tl.prepro.rgb_to_hsv(image)
    
    tl.vis.save_image(xx, r"img_result.jpg")

    #对图像文件夹里的图像随机高斯模糊
    import cv2,os
    import random
    def blur_img(imgName):
    ori_img = cv2.imread(imgName)
    size= random.choice((9,11,13,17))#高斯核
    kernel_size = (size, size)
    image = cv2.GaussianBlur(ori_img, ksize=kernel_size, sigmaX=0, sigmaY=0)
    cv2.imwrite(imgName, image)
    def random_blur(fileDir):
    pathDir = os.listdir(fileDir)
    filenumber = len(pathDir)
    rate = 0.7 # 自定义抽取图片的比例
    picknumber = int(filenumber * rate)
    sample = random.sample(pathDir, picknumber)
    count=0
    for name in sample:
        namepath=os.path.join(fileDir,name)
        blur_img(namepath)
        count+=1
        print(name+":"+str(count)+"/"+str(picknumber))

    import cv2 as cv
    import numpy as np 
    #模糊
    def blur_demo(image):
      dst = cv2.blur(image, (15, 1))
      cv2.imshow("blur_demo", dst)
    #锐化
    def sharpen_demo(image):
      kernel = np.array([[0, -1, 0], [-1, 5, -1], [0, -1, 0]], np.float32) 
      dst = cv.filter2D(image, -1, kernel=kernel)
      cv.imshow("custom_blur_demo", dst)
    src = cv2.imread("3.jpg")
    cv2.namedWindow("input image", cv.WINDOW_AUTOSIZE)
    cv2.imshow("input image", src)
    blur_demo(src)
    cv2.waitKey(0)
    cv2.destroyAllWindows()

    #图像融合
    import cv2
    import numpy as np
    import PIL.Image as Image
    # 统一图像大小
    im= cv2.imread(r"1.jpg")
    obj= cv2.imread(r"2.jpg")
    w=int(im.shape[0])
    h=int(im.shape[1])
    obj=cv2.resize(obj,(h,w),interpolation=cv2.INTER_CUBIC)
    # 生成全白的mask
    mask = 255 * np.ones(obj.shape, obj.dtype)
    width, height, channels = im.shape
    center = (height//2, width//2)
    # cv2.NORMAL_CLONE代表融合的模式，可以比较 cv2.NORMAL_CLONE和cv2.MIXED_CLONE的差别
    normal_clone = cv2.seamlessClone(obj, im, mask, center, cv2.NORMAL_CLONE)
    mixed_clone = cv2.seamlessClone(obj, im, mask, center, cv2.MIXED_CLONE)
    # Write results
    cv2.imwrite(r"C:\Users\An\Desktop\imgtest\2normal-clone.jpg", normal_clone)
    cv2.imwrite(r"C:\Users\An\Desktop\imgtest\2mixed-clone.jpg", mixed_clone)

    #图像随机擦除
    def transfrom(img,probability=1,sl=0.01, sh=0.03, r1=0.2, mean=(255,255,255)):
    if random.uniform(0, 1) > probability:
        return img
    area = img.shape[0] * img.shape[1]
    #面积
    target_area = random.uniform(sl, sh) * area
    #长宽比
    aspect_ratio = random.uniform(0.1,r1)
    h = int(round(math.sqrt(target_area * aspect_ratio)))
    w = int(round(math.sqrt(target_area /aspect_ratio)))
    if w < img.shape[1] and h < img.shape[0]:
        x1 = random.randint(0, img.shape[0] - h)
        y1 = random.randint(0, img.shape[1] - w)
        if img.shape[2] == 3:
            for i in range(x1,x1 + h):
                for j in range(y1,y1+w):
                    img[i,j,:]=mean
    
    return img