python_skimage

skimage

start

from skimage import io,data

chelsea = data.chelsea()

• 从skimage的data图片库数据库中导入一张图片

io.imshow(chelsea)

from skimage import color

gray_cat = color.rgb2gray(chelsea)
io.imshow(gray_cat)

chelsea.shape, gray_cat.shape

• ((300, 451, 3), (300, 451))
• 由三通道变成了单通道

hsv_cat = color.rgb2hsv(chelsea)
io.imshow(hsv_cat)

chelsea[0,0], hsv_cat[0,0]

• (array([143, 120, 104], dtype=uint8),
  array([0.06837607, 0.27272727, 0.56078431]))
• 直接对三通道分别做处理
• 取值范围由0~256 jpg 变成了0~1 png

from skimage import filters

二值化操作(灰度只有1,0)

t = filters.threshold_otsu(gray_cat)
for i in range(300):
    for j in range(451):
        if gray_cat[i, j] <= t:
            gray_cat[i, j] = 0
        else:
            gray_cat[i, j] = 1
io.imshow(gray_cat)

• threshold_otsu(gray_cat)

Return threshold value based on Otsu’s method.
根据Otsu方法返回一个值

• 图像需是单通道
  

验证码

cap = io.imread('capcha.jpg')
io.imshow(cap)

cap = color.rgb2gray(cap)   # 单通道
w,h = cap.shape
for i in range(w):
    for j in range(h):
        if cap[i, j] <= 0.1:    # 找到北京噪声
            cap[i, j] = 1      # 将噪声设成白色
io.imshow(cap)

t = filters.threshold_otsu(cap)
display(t)  #  0.464490456495098
for i in range(w):
    for j in range(h):
        if cap[i, j] <= t:
            cap[i, j] = 0
        else:
            cap[i, j] = 1
io.imshow(cap)

• t值太小,未完全区分
  

cap = io.imread('capcha.jpg')
cap = color.rgb2gray(cap)
w,h = cap.shape
for i in range(w):
    for j in range(h):
        if cap[i, j] <= 0.1:
            cap[i, j] = 1
t = filters.threshold_otsu(cap)
t = (t+1)/2
t = (t+1)/2
display(t)   # 0.8661226141237746
for i in range(w):
    for j in range(h):
        if cap[i, j] <= t:
            cap[i, j] = 0
        else:
            cap[i, j] = 1
io.imshow(cap)

• 手动增大一下