Numpy is a Python library similar to Matlab
It enables to work with multi-dimensional array without use of loops what is very important
since in Python, using loops for processing objects like images is very slow.

numpy is typically imported into Python as follows:
import numpy as np

numpy array can be created from analogical structure of Python lists and specified type:
a = np.array([[1,2],[3,4]],np.float32)
a
print(a)

and can be indexes as array
a[0,1]
a[-1,-1]

row can be accessed by
a[1]

column
a[:,1]

dimensions can be got
a.shape
number of rows is a.shape[0]
number of columns is a.shape[1]

each array has a type (np.int, np.uint8, np.float32)
a.dtype

and the type can be modified:
b = np.asarray(a,np.int)
b
b = a.astype(np.int)
b

array can be created also from array
c = np.array([a,b,b])
c 
array can be reshaped
d = a.reshape((1,4,1))
d
the easiest reshape is flattening
d = np.reshape(a,-1)
d
expand dimension
np.expand_dims(a,axis=2)

array can be created from zeros, ones, ones on diagonal, 
np.zeros((3,3),np.float)
np.ones((3,3),np.int)
np.eye(3)
from a constant
np.full((3,3),3)
as a sequence
v = np.arange(5)
randomly
np.random.uniform(0,1,(3,3))

sharing of array 
b = a 
b[0,0]=0
a[0,0]

copy of array
b = np.copy(a)  or  b = a.copy()
b[1,1]=1
a[1,1]

add, subtract, multiply, divide by constant
c = a + 2
c = a - 2
c = a * 2
c = a / 2
square each item
a ** 2
np.power(a,2)
exponential
2 ** a
np.exp(a*np.log(2)) 
binary logarithm
np.log(a) / np.log(2)
logarithm from 0 is -inf

add, subtract, multiply, divide by array
c = a + b
c = a - b
c = a * b
c = a / b
division by zero is nan

boolean arrays
a > 1
values = a[a > 1]
b[a == 2] = 4 
replace of nan by zero
c[c!=c]=0

scalar multiplication
np.dot(v,v)

transposition 
v = np.array([v])
v
v.T
np.transpose(v)

matrix multiplication
v @ v.T
v.T @ v
a @ a
np.dot(a,a)
np.matmul(a,a)

maximum, minimum
np.max(a)
np.min(a)

maximum in rows
np.max(a,axis=1)
maximum in columns
np.max(a,axis=0)

index of maximum
np.argmax(a,axis=0)

sum
a.sum()
np.sum(a)

average
v.mean()
np.mean(v)
deviation 
v.std()
normalization
(v-v.mean())/v.std()

matrix rotation
np.roll(a,1,axis=0)

apply boundary or low and high values
np.clip(a,1,2)

concatenation
np.concatenate((a,a), axis=0)
np.hstack((a,a))
np.vstack((a,a))

stack
np.array([a,a])

numpy has many parts: linalg (linear algebra), fft (Furier transform), ... 

e.g. norm of vector is np.linalg.norm(v)

opencv (cv2) is python library for computer vision
gray image is represented as 2D numpy array (rows,columns)
gray = cv2.imread("filename.png",cv2.IMREAD_GRAYSCALE)
color image is represented as 3D numpy arrays (rows,columns,channels)
bgr = cv2.imread("filename.png",cv2.IMREAD_COLOR)

image can be shown by
cv2.imshow('title',bgr)
cv2.waitKey(0)