计算机视觉图像处理

现实世界中的物体被映射到二维像素的网格中，这就是图像的形成方式。

图像的形成

图像可能包含 ：1、颜色。2、形状细节。3、光照条件。4、相机与物体的距离。

让我们首先尝试理解灰度图像。灰度图像表示为范围从0到255的像素值的网格。

灰度图像

彩色图像稍微复杂一点，它包含三个彩色通道。每个通道的值范围为0到255。

1、红色通道

2、绿色通道

3、蓝色通道

我们知道图像是像素值的网格，每个像素值的范围从0到255。我们将通过指定特定网格的行和列来了解这些值。让我们使用带有openCV库的python来实现这一点。

# Import openCV,matplotlib and numpy.
import cv2
import matplotlib.pyplot as plt
import numpy as np
img = cv2.imread('location/pic.jpg')
# Let's specify each row and column value as follows
x=10
y=10 
''' You will get 3 values for colour images [blue_color_values,green_color_values,red_color_values].each values represents corresponding B G and R Values,if the image is grayscale you will get a single value instead of three. '''
print(img[x,y])
''' Getting dimension of your image. (height, width,
number_of_color_channels). '''
print(img.shape)

现在做一些初级准备，如读取、复制Numpy数组中的图像、添加标题、将一种格式转换为另一种格式以及使用openCV python库绘制/显示或显示图像。

# importing(bringing-in) the essential libraries
# An Open-source library for Computer Vision and image processings, hence the name openCV.
import cv2
# To show the image/graphs etc.
import matplotlib.pyplot as plt
# Numpy helps us perform matrix (rows and columns of number) operation on images. Since, images are rows and columns of numbers.
import numpy as np
#The output of plotting commands is displayed inline within frontends like the Jupyter notebook.
#%matplotlib inline #uncomment %matplotlib inline while using jupyter.
# Reading in an image
img = cv2.imread('images.jpg')
# Always make a copy while operating on your image, so that your original image doesn't get affected.
img1 = np.copy(img)
#openCV reads/brings the image in BGR format (blue channel first, then green and finally red).
#If you plot/show your image using plt.imshow('img1'), it will appear weired. 
# Finally plot/show the image
plt.title('weired looking image in BGR format')
plt.imshow(img1)
# Reading in an image
img = cv2.imread('images.jpg')
# Always make a copy while operating on your image, so that your original image doesn't get affected.
img1 = np.copy(img)
#openCV reads/brings the image in BGR format (blue channel first, then green and finally red).
#If you plot/show your image using plt.imshow('img1'), it will appear weired. 
# Finally plot/show the image
plt.title('weired looking image in BGR format')
plt.imshow(img1)
#So you need to bring it to RGB format.By using cv2.cvtColor(img,cv2.COLOR_BGR2RGB).
img_bgr =np.copy(img)
img_rgb=cv2.cvtColor(img,cv2.COLOR_BGR2RGB)
plt.title('origional image in RGB format')
plt.imshow(img_rgb)

阈值化将灰度图像转换为二进制图像，其中将两个边界/级别分配给低于或高于指定阈值/边界的像素。

在创建蒙版时，我们将图像中的一些像素值设置为零或其他值。

# Thesholding a grayscale image.
# Create a new copy of image.
img_rgb1=np.copy(img_rgb)
# Convert it to grayscale.
img_gray=cv2.cvtColor(img_rgb1,cv2.COLOR_RGB2GRAY)
# Define upper and lower threshold values for grayscale image.
lower_values=np.array([70])
upper_values=np.array([250])
# create mask.
mask = cv2.inRange(img_gray,lower_values,upper_values)
#show/display it.
#cmap parameter refers to colour mapping, 
#which is a function that maps the colours of one image to the colours of another image.
plt.imshow(mask,cmap='gray')

我们可以使用不同的颜色值进行阈值处理，因为每种颜色我们需要3个不同的值。

# Thresholding a color image.
img_rgb1 = np.copy(img_rgb)
lower_values=np.array([70,70,10])
upper_values=np.array([250,250,250])
#creating mask
mask = cv2.inRange(img_rgb1,lower_values,upper_values)
#cmap parameter refers to colour mapping, 
#which is a function that maps the colours of one image to the colours of another image.
plt.imshow(mask,cmap='gray')

不同的颜色空间

RGB颜色空间中不同光照条件下物体的颜色有所不同。在HSV颜色空间中我们可以避免这样的问题。

让我们看看RGB和HSV颜色空间的对比。

在不同的HSV光照条件下，色相值保持相对恒定。而值在变化。

假设我们有一个蓝色背景的披萨。我们需要为它创建一个蒙版。在RGB颜色空间中做起来相当容易。

img_copy = np.copy(image)
# Do the above step as given in other codes blocks.
lower = np.array([0,0,230])
upper = np.array([50,50,255])
mask = cv2.inRange(img_copy, lower, upper)
plt.imshow(mask, cmap='gray')

img_copy = np.copy(img_rgb1)
# Do the above step as given in other codes blocks.
lower = np.array([0,0,225])
upper = np.array([250,250,255])
mask = cv2.inRange(img_copy, lower, upper)
plt.imshow(mask, cmap='gray')

我们可以看出区别，阴影部分也考虑在黑色部分。这个蒙版比之前的要糟糕。

我们来比较一下

u_img = cv2.imread('uniform.png')
# Always make a copy while operating on your image, so that your original image doesn't get affected.
img_u = np.copy(u_img)
img_rgb = cv2.cvtColor(img_u,cv2.COLOR_BGR2RGB)
img_hsv = cv2.cvtColor(img_rgb,cv2.COLOR_RGB2HSV)
# Converting rgb to hsv color space.
img_hsv=cv2.cvtColor(u_img,cv2.COLOR_BGR2HSV)
# selecting all rows,columns and hue.
h=img_hsv[:,:,0]
# selecting all rows,columns and saturation.
s=img_hsv[:,:,1]
# selecting all rows,columns and values.
v=img_hsv[:,:,2]
#ploting multiple images using subplot
f,(ax1,ax2,ax3,ax4,ax5) = plt.subplots(1,5,figsize=(20,10))
ax1.imshow(img_rgb,cmap='gray')
ax2.imshow(img_hsv,cmap='gray')
ax3.imshow(h,cmap='gray')
ax4.imshow(s,cmap='gray')
ax5.imshow(v,cmap='gray')

输出1

un_img = cv2.imread('un_uniform.png')
# Always make a copy while operating on your image, so that your original image doesn't get affected.
img_un = np.copy(un_img)
img_rgb1 = cv2.cvtColor(img_un,cv2.COLOR_BGR2RGB)
img_hsv1 = cv2.cvtColor(img_rgb1,cv2.COLOR_RGB2HSV)
# Converting rgb to hsv color space.
img_hsv1=cv2.cvtColor(un_img,cv2.COLOR_BGR2HSV)
# selecting all rows,columns and hue.
h1=img_hsv1[:,:,0]
# selecting all rows,columns and saturation.
s1=img_hsv1[:,:,1]
# selecting all rows,columns and values.
v1=img_hsv1[:,:,2]
#ploting multiple images using subplot
f,(ax1,ax2,ax3,ax4,ax5) = plt.subplots(1,5,figsize=(20,10))
ax1.imshow(img_rgb1,cmap='gray')
ax2.imshow(img_hsv1,cmap='gray')
ax3.imshow(h1,cmap='gray')
ax4.imshow(s1,cmap='gray')
ax5.imshow(v1,cmap='gray')

输出2

在两个输出中，我们可以看到中间图像看起来相同。这是色彩，我们可以使用这些信息进行屏蔽或其他操作。

# Caution : Hue is angle ranges from 0 to 180.
lower_values=np.array([90,10,113])
upper_values=np.array([180,255,255])
mask2 = cv2.inRange(img_hsv1,lower_values,upper_values)
#cmap parameter refers to colour mapping, 
#which is a function that maps the colours of one image to the colours of another image.
plt.imshow(mask2,cmap='gray')

更好的蒙版使用使用色调颜色空间