图像增强

发表于

Task

Reference

Imgaug

构造数据增强器

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
from imgaug import augmenters as iaa

sometimes = lambda aug: iaa.Sometimes(0.5, aug) #建立lambda表达式，

seq = iaa.Sequential(
    [
        sometimes(iaa.Affine(  
            scale={"x": (0.8, 1.2), "y": (0.8, 1.2)},#图像缩放为80%到120%之间
            rotate=(-3, 3),   #旋转±3度之间
            order=[0, 1],   #使用最邻近差值或者双线性差值
        )),
        
        iaa.SomeOf((0, 5),
            [
                # 将部分图像进行超像素的表示。
                sometimes(
                    iaa.Superpixels(
                        p_replace=(0, 1.0),
                        n_segments=(20, 200)
                    )
                ),

                #用高斯模糊，均值模糊，中值模糊中的一种增强。注意OneOf的用法
                iaa.OneOf([
                    iaa.GaussianBlur((0, 3.0)),
                    iaa.AverageBlur(k=(2, 7)), # 核大小2~7之间，k=((5, 7), (1, 3))时，核高度5~7，宽度1~3
                    iaa.MedianBlur(k=(3, 11)),
                ]),


                # 加入高斯噪声
                iaa.AdditiveGaussianNoise(
                    loc=0, scale=(0.0, 0.05*255), per_channel=0.5
                ),

                # 每个像素随机加减-10到10之间的数
                iaa.Add((-10, 10), per_channel=0.5),


                # 将RGB变成灰度图然后乘alpha加在原图上
                iaa.Grayscale(alpha=(0.0, 1.0)),

            ],
            
            random_order=True # 随机的顺序把这些操作用在图像上
        )
    ],
    random_order=True # 随机的顺序把这些操作用在图像上
)

使用数据增强器：

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
file_list = os.listdir('1_aug/')
pre_path = ('1_aug/')

aug_gen_cnt = 2

for it in file_list:
    for _ in range(aug_gen_cnt):
        img1 = seq.augment_image(cv2.imread(pre_path + it))
        #img2 = seq.augment_image(cv2.imread(pre_path + it))
        
        name_pos = re.search('.jpg', file_list[0]).span()
        name_pre = it[:name_pos[0]]
        
        cv2.imwrite(pre_path + name_pre + '_aug_1.jpg', img1)
#         cv2.imwrite(pre_path + name_pre + '_aug_2.jpg', img2)

Opencv

  • 灰度图 To 颜色3通道图

    1
    2
    img1 = cv2.imread(it1, cv2.IMREAD_GRAYSCALE)
    img1 = cv2.cvtColor(img1, cv2.COLOR_GRAY2BGR)

  • 模糊去噪与梯度滤波处理边缘检测——拉普拉斯算子+中值模糊+反转正则化

    1
    2
    3
    4
    5
    6
    7
    8
    9
    10
    11
    12
    13
    def edge_aug(img):
        img_process_1 = cv2.medianBlur(img, 7)
        img_process_1_gray = cv2.cvtColor(img_process_1, cv2.COLOR_BGR2GRAY)
        cv2.Laplacian(img_process_1_gray, cv2.CV_8U, img_process_1_gray, ksize=5)
        normalizedInverse = (1.0 / 255) * (255 - img_process_1_gray)
        channels = cv2.split(img)
        
        for channel in channels:
            channel[:] = channel * normalizedInverse

        dst = cv2.merge(channels)
        
        return dst

  • Canny边缘检测

    1
    2
    3
    4
    5
    6
    7
    8
    9
    10
    11
    12
    13
    pre_path = 'rec_res/'
    tar_path = 'edge_enforce/'

    for img_path in os.listdir(pre_path):
        img = cv2.imread(pre_path + img_path)
        
    #     dst_temp = edge_aug(img)
        edges = cv2.Canny(img,100,200)
        
        cv2.imwrite(tar_path + img_path, edges)

    #plt.imshow(edges, cmap='gray')
    #plt.show()

  • 角点检测

    1
    2
    3
    4
    5
    6
    7
    img = cv2.imread('621669_7600000020025.jpeg')
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    gray = np.float32(gray)
    dst = cv2.cornerHarris(gray, 2, 23, 0.04)
    img[dst>0.01 * dst.max()] = [0, 0, 255]
    plt.imshow(img)
    plt.show()

  • Resize

    1
    2
    3
    4
    5
    6
    7
    8
    9
    10
    11
    12
    13
    14
    15
    16
    17
    18
    19
    20
    21
    22
    23
    24
    def resize_im(im, scale, max_scale=None):
        f = float(scale) / min(im.shape[0], im.shape[1])
        if max_scale != None and f * max(im.shape[0], im.shape[1]) > max_scale:
            f = float(max_scale) / max(im.shape[0], im.shape[1])
        return cv2.resize(im, None, None, fx=f, fy=f, interpolation=cv2.INTER_LINEAR), f

    img = cv2.imread('621669_7600000020025.jpeg')
    SCALE=900  # 600
    MAX_SCALE=1500  # 1200
    TEXT_PROPOSALS_WIDTH=0  # 16
    MIN_NUM_PROPOSALS=0  # 2
    MIN_RATIO=0.01  # 0.5
    LINE_MIN_SCORE=0.6  # 0.9
    MAX_HORIZONTAL_GAP=30  # 50
    TEXT_PROPOSALS_MIN_SCORE=0.7  # 0.7
    TEXT_PROPOSALS_NMS_THRESH=0.3  # 0.2
    TEXT_LINE_NMS_THRESH=0.3
    MIN_V_OVERLAPS=0.6  # 0.7
    MIN_SIZE_SIM=0.6  # 0.7

    img, scale = resize_im(img, scale=SCALE,
                               max_scale=MAX_SCALE)
    plt.imshow(img_temp)
    plt.show()

Pillow

  • 翻转

    1
    2
    3
    4
    5
    from PIL import Image,ImageEnhance,ImageFilter,ImageOps

    img = Image.open(x)
    img_flip_left_right = img.transpose(Image.FLIP_LEFT_RIGHT)
    img_flip_top_bottom = img.transpose(Image.FLIP_TOP_BOTTOM)

  • 旋转

    1
    2
    3
    4
    5
    6
    7
    8
    from PIL import Image,ImageEnhance,ImageFilter,ImageOps

    img = Image.open(x)
    img_rotate_90 = img.transpose(Image.ROTATE_90)
    img_rotate_180 = img.transpose(Image.ROTATE_180)
    img_rotate_270 = img.transpose(Image.ROTATE_270)
    img_rotate_90_left = img_flip_left_right.transpose(Image.ROTATE_90)
    img_rotate_270_left = img_flip_left_right.transpose(Image.ROTATE_270)

  • 3.亮度

    1
    2
    3
    4
    5
    6
    from PIL import Image,ImageEnhance,ImageFilter,ImageOps

    img = Image.open(x)
    enh_bri = ImageEnhance.Brightness(img)
    brightness = 1.5
    image_brightened = enh_bri.enhance(brightness)

  • 色彩

    1
    2
    3
    4
    5
    6
    from PIL import Image,ImageEnhance,ImageFilter,ImageOps

    img = Image.open(x)
    enh_col = ImageEnhance.Color(img)
    color = 1.5
    image_colored = enh_col.enhance(color)

  • 对比度

    1
    2
    3
    4
    5
    6
    from PIL import Image,ImageEnhance,ImageFilter,ImageOps

    img = Image.open(x)
    enh_con = ImageEnhance.Contrast(img)
    contrast = 1.5
    image_contrasted = enh_con.enhance(contrast)

  • 锐度

    1
    2
    3
    4
    5
    6
    from PIL import Image,ImageEnhance,ImageFilter,ImageOps

    img = Image.open(x) 
    enh_sha = ImageEnhance.Sharpness(img)
    sharpness = 3.0
    image_sharped = enh_sha.enhance(sharpness)

Skimage

  • 加高斯噪声

    1
    2
    3
    4
    5
    6
    7
    8
    9
    10
    11
    12
    from skimage.util import random_noise
    import cv2
    # 加高斯噪声
    def addNoise(img):
        '''
        注意：输出的像素是[0,1]之间,所以乘以5得到[0,255]之间
        '''
        return random_noise(img, mode='gaussian', seed=13, clip=True)*255
        
    cv_image = cv2.imread(x)
    # 高斯噪声
    gau_image = addNoise(cv_image)

  • 随机改变亮度

    1
    2
    3
    4
    5
    6
    7
    8
    9
    10
    11
    12
    13
    import cv2
    import random
    from skimage import exposure

    def changeLight(img):
        rate = random.uniform(0.5, 1.5)
        # print(rate)
        img = exposure.adjust_gamma(img, rate) #大于1为调暗，小于1为调亮;1.05
        return img
        
    cv_image = cv2.imread(x)
    # 随机改变
    light = changeLight(cv_image)

Other Skill

  • 搜寻图片名字

    1
    2
    3
    4
    5
    6
    re.search和re.match区别：match只匹配开头
    - match        
    #re.match('row/0/', 'row/0/_001a_0.png').span()
    #（0， 6）
    - search
    #shutil.copyfile(it, tar_pre_path + it[re.search(ori_pre_path, it).span()[1]:])

  • shutil复制图片

    1
    2
    3
    4
    5
    ori_pre_path = './0_aug/'
    tar_pre_path = './images/'

    for it in  glob.glob(ori_pre_path + '*.jpg'):
        shutil.copyfile(it, tar_pre_path + it[re.search(ori_pre_path, it).span()[1]:])

  • 图像拼接&裁剪

    1
    2
    3
    4
    5
    6
    ①
    for it in test_list[1:4]:
        img_res = np.hstack([img_res, it])
        
    ②
    img_cut = test_list[4][:, :test_list[4].shape[1] // 4 * 3, :]

  • Keras—data_generater的写法

    【Github】刘扬农作物病虫害检测

TODO

  • pydicom (医学，dcm)
  • SimpleITK (医学，dcm)