How to get barcode from image?

Question

I need to get the information in the below barcode with the Python pyzbar library, but it does not recognize it. Should I make any improvement before using pyzbar?

BarCode

this is the code:

from pyzbar.pyzbar import decode
import cv2

    def barcodeReader(image):
        gray_img = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
        barcodes = decode(gray_img)

    barcode = barcodeReader("My_image")
    print (barcode)

Result: []

You could try applying a sharpening filter but its probably not getting recognized because your image is too damaged — nathancy, Dec 02 '19 at 22:13

HansHirse · Accepted Answer · 2019-12-03T11:27:04.120

You could try to reconstruct the barcode by:

Inverse binarizing the image with cv2.threshold, such that you get white lines on black background.
Counting all non-zero pixels along the rows using np.count_nonzero.
Getting all indices, where the count exceeds a pre-defined threshold, let's say 100 here.
On a new, all white image, drawing black lines at the found indices.

Here's some code:

import cv2
import numpy as np
from skimage import io      # Only needed for web grabbing images, use cv2.imread for local images

# Read image from web, convert to grayscale, and inverse binary threshold
image = cv2.cvtColor(io.imread('https://i.stack.imgur.com/D8Jk7.jpg'), cv2.COLOR_RGB2GRAY)
_, image_thr = cv2.threshold(image, 128, 255, cv2.THRESH_BINARY_INV)

# Count non-zero pixels along the rows; get indices, where count exceeds certain threshold (here: 100)
row_nz = np.count_nonzero(image_thr, axis=0)
idx = np.argwhere(row_nz > 100)

# Generate new image, draw lines at found indices
image_new = np.ones_like(image_thr) * 255
image_new[35:175, idx] = 0

cv2.imshow('image_thr', image_thr)
cv2.imshow('image_new', image_new)
cv2.waitKey(0)
cv2.destroyAllWindows()

The inverse binarized image:

The reconstructed image:

I'm not sure, if the result is a valid barcode. To improve the solution you could get rid of the numbers beforehand. Also, play around with the threshold.

Hope that helps!

@AndresMarin That's a pity! But as nathancy already stated in his comment, your image then might be too broken for a proper reconstruction. — HansHirse, Dec 04 '19 at 05:02
since this came up in another question: this method is equivalent to integrating/summing along columns and then thresholding. the idea is to suppress noise using integration. — Christoph Rackwitz, Dec 23 '20 at 21:46

score 0 · Answer 2 · edited Jun 09 '21 at 13:29

You can follow below approach:

Using morphological operation detect vertical lines and stored the xmin ymin, xmax and ymax of vertical image.
sort all xmin values and grouped them based distance.
do same excercise ymin and ymax and grouped them.
consider smallest pixels values from larger group of xmin and ymin larger group respectively.
consider largest values from larger group of xmax and ymax larger grop respectively.
you will get exact xmin,ymin,xmax,ymax of barcode.

How to get barcode from image?

2 Answers2