How to extract characters from images where their positions are known? - image

Question

I have a set of png images of 300dpi . Each image is full of text (not handwritten), digits (not handwritten).
l want to extract each character and save it in a different image.
For each character in the image l have its position stored in csv file.
For instance in image1.png for a given character “k” l have its position :
“k”=[left=656, right=736,top=144,down= 286]
Is there any python library which allows to do that ?. As input l have the images (png format) and csv file that contains the position of each character of each images.
after executing the code l stack at this line :
img_charac=img[int(coords[2]):int(coords[3]),int(coords[0]):int(coords[1])]
l got the following error:
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
TypeError: 'NoneType' object has no attribute '__getitem__'

Answer 1

So if I understood correctly, this has nothing to do with image processing, just file opening, image cropping and saving.
With a csv file looking like ,
an input image looking like
I get results like
import cv2
import numpy as np
import csv
path_csv= #path to your csv
#stock coordinates of characters from your csv in numpy array
npa=np.genfromtxt(path_csv+"cs.csv", delimiter=',',skip_header=1,usecols=(1,2,3,4))
nb_charac=len(npa[:, 0]) #number of characters
#stock the actual letters of your csv in an array
characs=[]
cpt=0
#take characters
f = open(path_csv+"cs.csv", 'rt')
reader = csv.reader(f)
for row in reader:
if cpt>=1: #skip header
characs.append(str(row[0]))
cpt+=1
#open your image
path_image= #path to your image
img=cv2.imread(path_image+"yourimagename.png")
path_save= #path you want to save to
#for every line on your csv,
for i in range(nb_charac):
#get coordinates
coords=npa[i,:]
charac=characs[i]
#actual cropping of the image (easy with numpy)
img_charac=img[int(coords[2]):int(coords[3]),int(coords[0]):int(coords[1])]
#saving the image
cv2.imwrite(path_save+"carac"+str(i)+"_"+str(charac)+".png",img_charac)
This is sort of quick and dirty, the csv opening is a bit messy for example (you could get all the info with one opening and converting), and should be adapted to your csv file anyway.