OpenCV image processing ---- image binarization

图像二值化（ Image Binarization）It is to set the grayscale value of the pixels on the image to 0 or 255, which is the process of presenting an obvious black and white effect to the entire image.

---

The principle of binarization

import cv2

img = cv2.imread('img/lena.jpg')
# 转为灰度图
new_img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
height, width = new_img.shape[0:2]

# 设置阈值
thresh = 60

# 遍历每一个像素点
for row in range(height):
    for col in range(width):
        # 获取到灰度值
        gray = new_img[row, col]
        # 如果灰度值高于阈值 就等于255最大值
        if gray > thresh:
            new_img[row, col] = 255
        # 如果小于阈值，就直接改为0
        elif gray < thresh:
            new_img[row, col] = 0

cv2.imshow('img', new_img)
cv2.waitKey()

---

Image binarization API provided by OpenCV

threshold()方法参数：

1. image matrix
2. threshold
3. maximum value in the picture
4. binarization method

The way of binarization:

THRESH_BINARY	Above the threshold is changed to 255, below the threshold is changed to 0
THRESH_BINARY_INV	Above the threshold is changed to 0, below the threshold is changed to 255
THRESH_TRUNC	Truncate, change above threshold to threshold, max invalidation
THRESH_TOZERO	Above the threshold does not change, below the threshold changes to 0
THRESH_TOZERO_INV	If it is higher than the threshold, it should be 0, and if it is lower than the threshold, it will not change

import cv2

img = cv2.imread('img/lena.jpg', cv2.IMREAD_GRAYSCALE)

thresh, new_img = cv2.threshold(img, 60, 255, cv2.THRESH_BINARY)

print(thresh)
cv2.imshow('img', img)
cv2.imshow('NEW_IMG', new_img)
cv2.waitKey()

---

adaptive threshold

Use a global value as the threshold. But this may not be good in all cases. If the image has different lighting conditions in different areas. In this case, Adaptive Thresholding can help. Here, the algorithm thresholds a pixel based on a small area around it. Therefore, we obtain different thresholds for different regions of the same image, which gives better results for images with different lighting.

adaptlive()方法参数：

1. image matrix
2. Image grayscale maximum
3. Method to Calculate Threshold
4. threshold type
5. processing block size
6. The constant C used by the algorithm

• cv2.ADAPTIVE_THRESH_MEAN_C: The threshold is the constant mean C subtracted from the neighborhood.
• cv2.ADAPTIVE_THRESH_GAUSSIAN_C: The threshold is the Gaussian weighted sum of the neighborhood values ​​minus the constant C.

import cv2

img = cv2.imread('img/lena.jpg', cv2.IMREAD_GRAYSCALE)

thresh_img = cv2.adaptiveThreshold(img, 255, cv2.ADAPTIVE_THRESH_MEAN_C, cv2.THRESH_BINARY, 11, 5)

cv2.imshow('thresh_img', thresh_img)
cv2.waitKey()

Otsu algorithm (maximum between-class variance method)

图像分割中阈值选取的最佳算法

threshold(gaussian_img, 0, 255, cv2.THRESH_BINARY | cv2.THRESH_OTSU)

import cv2

img = cv2.imread('img/lena.jpg', cv2.IMREAD_GRAYSCALE)

# 使用255的阈值进行二值化
ret, thresh_img = cv2.threshold(img, 255, 255, cv2.THRESH_BINARY)
cv2.imshow('normal', thresh_img)

# 使用高斯滤波模糊图像  参数1： 图片矩阵  参数2：卷积核 参数3： 越大越模糊
gaussian_img = cv2.GaussianBlur(img, (5, 5), 0)
cv2.imshow('gaussian_img', gaussian_img)

# 使用大津算法0阈值二值化经过高斯滤波模糊后的图像
ret, thresh_img = cv2.threshold(gaussian_img, 0, 255, cv2.THRESH_BINARY | cv2.THRESH_OTSU)

cv2.imshow('otsu', thresh_img)

cv2.imshow('img', img)
cv2.waitKey()

Original image:

Use a threshold of 255 for binarization and the image is all black:

Blur an image using a Gaussian filter:

1. image matrix
2. convolution kernel
3. bigger and blurrier

Use the Otsu algorithm with a threshold of 0 to binarize the image blurred by Gaussian filtering: