模式识别与图像处理课程实验一：图像处理实验(颜色算子实验、Susan、Harris角点检测实验、 sobel边缘算子检测实验)

一、 实验内容

要求编写一个包含颜色算子，Susan，Harris，角点，sobel边缘算子的程。

二、 颜色算子实验

2.1、 提取红色

• 实验的程序如下

import numpy as np
import cv2 as cv

image = cv.imread("1.jpg")
image = image / np.ones([1, 1, 3]).astype(np.float32)
image = cv.cvtColor(image, cv.COLOR_BGR2RGB)
print(image.shape)

# 颜色算子
# red
redAdd = np.ones([1, 1, 3]).astype(np.float32)
redAdd[0, 0, 0] = 1.0
redAdd[0, 0, 1] = 0.5
redAdd[0, 0, 2] = 0.25

redSub = np.ones([1, 1, 3]).astype(np.float32)
redSub[0, 0, 0] = 0.25
redSub[0, 0, 1] = 0.5
redSub[0, 0, 2] = 1.0

image1 = np.mean(image * redAdd, 2)
image2 = np.mean(image * redSub, 2) + 100
imageRed = image1 / image2

redMax = np.max(imageRed)
redMin = np.min(imageRed)

imageRed = 255 * (imageRed - redMin) / (redMax - redMin)
cv.imwrite("1red.png", imageRed)

运行结果如下

• 实验原图
  
  • 实验结果图
    

2.2、 提取绿色

实验的程序如下

import numpy as np
import cv2 as cv

image = cv.imread("1.jpg")
image = image / np.ones([1, 1, 3]).astype(np.float32)
image = cv.cvtColor(image, cv.COLOR_BGR2RGB)
print(image.shape)

# green
greenAdd = np.ones([1, 1, 3]).astype(np.float32)
greenAdd[0, 0, 0] = 0.5
greenAdd[0, 0, 1] = 1.0
greenAdd[0, 0, 2] = 0.25

greenSub = np.ones([1, 1, 3]).astype(np.float32)
greenSub[0, 0, 0] = 0.5
greenSub[0, 0, 1] = 0.25
greenSub[0, 0, 2] = 1.0

image1 = np.mean(image * greenAdd, 2)
image2 = np.mean(image * greenSub, 2) + 100

imageGreen = image1 / image2
greenMax = np.max(imageGreen)
greenMin = np.min(imageGreen)

imageRed = 255 * (imageGreen - greenMin) / (greenMax - greenMin)
cv.imwrite("1green.png", imageRed)

运行结果如下

1. 实验原图
   

2. 实验结果图
   

2.3、 提取蓝色

• 实验的程序如下

import numpy as np
import cv2 as cv

image = cv.imread("1.jpg")
image = image / np.ones([1, 1, 3]).astype(np.float32)
image = cv.cvtColor(image, cv.COLOR_BGR2RGB)
print(image.shape)

# bule
buleAdd = np.ones([1, 1, 3]).astype(np.float32)
buleAdd[0, 0, 0] = 0.25
buleAdd[0, 0, 1] = 0.5
buleAdd[0, 0, 2] = 1.0

buleSub = np.ones([1, 1, 3]).astype(np.float32)
buleSub[0, 0, 0] = 1.0
buleSub[0, 0, 1] = 0.5
buleSub[0, 0, 2] = 0.25

image1 = np.mean(image * buleAdd, 2)
image2 = np.mean(image * buleSub, 2) + 100

imageBlue = image1 / image2
blueMax = np.max(imageBlue)
blueMin = np.min(imageBlue)

imageBlue = 255 * (imageBlue - blueMin) / (blueMax - blueMin)

cv.imwrite("1blue.png", imageBlue)

运行结果如下

1. 实验原图
   

2. 实验结果图

三、 Susan、Harris角点检测实验

3. 1、 实验程序

3.1.1、Susan角点检测

Susan角点检测程序如下

import numpy as np
import cv2 as cv

image = cv.imread("2.jpg")
image = np.mean(image, 2)

height = image.shape[0]
width = image.shape[1]
print(image.shape)

#susan 算子
radius = 5
imageSusan = np.zeros([height, width]).astype(np.float32)

for h in range(radius, height-radius):
    for w in range(radius, width-radius):
        numSmall = 0
        numLarge = 0
        numAll = 0

        for y in range(-radius, radius + 1):
            for x in range(-radius, radius+1):
                distance = np.sqrt(y**2 + x**2)

                if distance <= radius:
                    numAll += 1

                    if image[h + y, w + x] < image[h, w] - 27:
                        numSmall += 1
                    if image[h + y, w + x] > image[h, w] + 27:
                        numLarge += 1

        ratio = 1.0 * numSmall / numAll
        ratio2 = 1.0 * numLarge / numAll

        if ratio < 0.3:
            imageSusan[h, w] = 0.3 - ratio
        if ratio2 > 0.7:
            imageSusan[h, w] = ratio2 - 0.7

imageMax = np.max(imageSusan)
imageMin = np.min(imageSusan)

imageSusan = 255*(imageSusan - imageMin)/(imageMax - imageMin)
print(imageSusan.shape)

cv.imwrite("2.png", imageSusan)

• 运行结果如下

1. 实验原图
   

2. 实验结果图
   

3.1.2、Harris角点检测

• Harris角点检测程序如下

import cv2 as cv
import numpy as np
import matplotlib.pyplot as plt

# 读取图像
img = cv.imread('3.jpg')
lenna_img = cv.cvtColor(img, cv.COLOR_BGR2RGB)

# 图像转换成灰度图像
grayImage = cv.cvtColor(img, cv.COLOR_BGR2GRAY)
grayImage = np.float32(grayImage)

# Harris算子
harrisImage = cv.cornerHarris(grayImage, 2, 3, 0.04)

harrisImage = cv.dilate(harrisImage, None)

# 设置阈值
thresImage = 0.006 * harrisImage.max()

img[harrisImage > thresImage] = [255, 0, 0]

# 显示正常中文的标签
plt.rcParams['font.sans-serif'] = ['SimHei']

titles = [u'(a)原始图像', u'(b)Harris图像']
images = [lenna_img, img]

for i in range(2):
    plt.subplot(1, 2, i + 1), plt.imshow(images[i], 'gray')
    plt.title(titles[i])
    plt.xticks([]), plt.yticks([])
plt.show()

运行结果如下

四、 sobel边缘算子检测实验

4.1、sobel边缘算子检

• sobel边缘算子检程序如下

import numpy as np
import cv2

image = cv2.imread("3.jpg")
height = image.shape[0]
width = image.shape[1]

sobelResult = np.zeros([height - 2, width - 2, 1]).astype(np.float32)
sobelX = np.zeros([3, 3, 1]).astype(np.float32)
sobelY = np.zeros([3, 3, 1]).astype(np.float32)

sobelX[0, 0, 0] = -1
sobelX[1, 0, 0] = -2
sobelX[2, 0, 0] = -1
sobelX[0, 2, 0] = 1
sobelX[1, 2, 0] = 2
sobelX[2, 2, 0] = 1

sobelY[0, 0, 0] = -1
sobelY[0, 1, 0] = -2
sobelY[0, 2, 0] = -1
sobelY[2, 0, 0] = 1
sobelY[2, 1, 0] = 2
sobelY[2, 2, 0] = 1


for h in range(0, height - 3):
    for w in range(0, width - 3):
        #求方向梯度
        imageIncre = image[h:h + 3, w:w + 3]
        gradientX = np.sum(imageIncre * sobelX)
        gradientY = np.sum(imageIncre * sobelY)
        gradient = np.sqrt(gradientX**2 + gradientY**2)
        sobelResult[h, w, 0] = gradient


imageMax = np.max(sobelResult)
imageMin = np.min(sobelResult)

sobelResult = 255*(sobelResult - imageMin) / (imageMax - imageMin)
cv2.imwrite("3.png", sobelResult)

2、 运行结果如下

1. 实验原图
   

2. 实验结果图
   

五、 实验总结

• 1、 掌握了编写含颜色算子图像处理、Susan与Harris角点图像检测、sobel边缘算子图像检测的程序编写方法。
• 2、 通过实验、对于边缘检测算子与角点检测算子有了进一步的掌握。