I set myself the task of recognizing passports, but I can’t completely recognize all areas. Tell me, what can help? Used a different filtering and canny algorithm, but something is missing. Код не может распознать серию и номер документа, а также мелкие символы, иногда не может распознать имя или фамилию совсем....
# import the necessary packages
from PIL import Image
import pytesseract
import argparse
import cv2
import os
import numpy as np
# построить разбор аргументов и разбор аргументов
ap = argparse.ArgumentParser()
ap.add_argument("-i", "--image" )
ap.add_argument("-p", "--preprocess", type=str, default="thresh")
args = vars(ap.parse_args())
# загрузить пример изображения и преобразовать его в оттенки серого
image = cv2.imread ("pt.jpg")
gray = cv2.cvtColor (image, cv2.COLOR_BGR2GRAY)
gray = cv2.Canny(image,300,300,apertureSize = 3)
# check to see if we should apply thresholding to preprocess the
# image
if args["preprocess"] == "thresh":
gray = cv2.threshold (gray, 0, 255, cv2.THRESH_BINARY | cv2.THRESH_OTSU)[1]
# make a check to see if median blurring should be done to remove
# noise
elif args["preprocess"] == "blur":
gray = cv2.medianBlur (gray, 3)
# write the grayscale image to disk as a temporary file so we can
# apply OCR to it
filename = "{}.png".format (os.getpid ())
cv2.imwrite (filename, gray)
# load the image as a PIL/Pillow image, apply OCR, and then delete
# the temporary file
pytesseract.pytesseract.tesseract_cmd = r'C:\Program Files\Tesseract-OCR\tesseract.exe'
text = pytesseract.image_to_string (image, lang = 'rus+eng')
os.remove (filename)
print (text)
os.system('python gon.py > test.txt') # doc output file
# show the output images
cv2.imshow ("Image", image)
cv2.imshow ("Output", gray)
cv2.waitKey (0)
It's easier (and faster) for Tesseract to recognize text when you provide it only with the regions that contain the text you want to interpret, in your case the big, black letters in the middle, for example:
I'm referring to running Tesseract only in the regions in green. Since the document's structure is predictable, you could easily find these regions as follows:
binarize and invert the image (black = empty)
use opencv's connectedComponentsWithStats() function to get a list of all connected components with their positions and size
you can hardcode thresholds to filter only the characters you want, or get a histogram of areas and use statistics to define thresholds dinamically
on the remaining connected components use morphological operations (e.g. dilation with a horizontal kernel) to connect letters together horizontally
get the bounding box of the final connected components
Optional: postprocessing will work much better in these isolated boxes
Feed each bounding box as a separate Mat to tesseract, it will greatly simplify the problem.
Related
I have been trying to segment biological cells in an image using watershed algorithm. I found an excellent article on pyimagesearch which clearly gives an overview of the algorithm and its implementation in python. The code uses both opencv and scikit-image for processing the image.
My goal is to convert the whole code into pure opencv. But the issues is that there's a function called scipy.feature.peak_local_max in scikit-image which does the job of finding local peaks in an image very efficiently. I couldn't find or devise such function in OpenCV.
Original Code(I have documented this snippet according to my understanding, please correct if am wrong):
import the necessary packages
from skimage.feature import peak_local_max
from skimage.morphology import watershed
from scipy import ndimage
import numpy as np
import argparse
import imutils
import cv2
from matplotlib import pyplot as plt
# load the image and perform pyramid mean shift filtering
# to aid the thresholding step
image = cv2.imread("test2.png")
shifted = cv2.pyrMeanShiftFiltering(image, 21, 51)
# Apply grayscale
gray = cv2.cvtColor(shifted, cv2.COLOR_BGR2GRAY)
# Convert to binary
thresh = cv2.threshold(gray, 0, 255,cv2.THRESH_BINARY_INV | cv2.THRESH_OTSU)[1]
# Watershed starts from here
# compute the exact Euclidean distance from every binary
# pixel to the nearest zero pixel, then find peaks in this
# distance map
D = ndimage.distance_transform_edt(thresh)
localMax = peak_local_max(D, indices=False, min_distance=10,labels=thresh)
# perform a connected component analysis on the local peaks,
# using 8-connectivity, then appy the Watershed algorithm
markers = ndimage.label(localMax, structure=np.ones((3, 3)))[0]
# Apply segmentation
labels = watershed(-D, markers, mask=thresh)
print("[INFO] {} unique segments found".format(len(np.unique(labels)) - 1))
cv2.imwrite("labels.png",labels)
# Contouring
for label in np.unique(labels):
# if the label is zero, we are examining the 'background'
# so simply ignore it
if label == 0:
continue
# otherwise, allocate memory for the label region and draw
# it on the mask
mask = np.zeros(gray.shape, dtype="uint8")
mask[labels == label] = 255
cnts = cv2.findContours(mask.copy(), cv2.RETR_EXTERNAL,cv2.CHAIN_APPROX_SIMPLE)
cnts = imutils.grab_contours(cnts)
c = max(cnts, key=cv2.contourArea)
# draw a circle enclosing the object
peri = cv2.arcLength(c, True)
approx = cv2.approxPolyDP(c, 0.018 * peri, True)
cv2.drawContours(image, [approx], -1, (0,0,255), 2)
cv2.imwrite("output.jpg",image)
Pure OpenCV Code till finding distance map:
# import the necessary packages
import numpy as np
import cv2
# load the image and perform pyramid mean shift filtering
# to aid the thresholding step
image = cv2.imread("1.png")
shifted = cv2.pyrMeanShiftFiltering(image, 21, 51)
# Apply grayscale
gray = cv2.cvtColor(shifted, cv2.COLOR_BGR2GRAY)
# Convert to binary
thresh = cv2.threshold(gray, 0, 255,cv2.THRESH_BINARY_INV | cv2.THRESH_OTSU)[1]
# Watershed starts from here
# compute the exact Euclidean distance from every binary
# pixel to the nearest zero pixel, then find peaks in this
# distance map
D = cv2.distanceTransform(thresh,cv2.DIST_L2,0)
The point till D, both the original code and the pure opencv code which I have tried have exactly the same outputs, the issue is I dont exactly have a clear idea on how to implement peak_local_max in opencv which would give identical result as scikit's function.
It would be really helpful if someone who has relavent knowledge could explain how this function works in finding those peaks in such a fine grained manner.
Input Image:
Peak Local max output in scikit-image(BGR format image):
Required output:
I need assistance with a ML project I am currently trying to create.
I receive a lot of invoices from a lot of different suppliers - all in their own unique layout. I need to extract 3 key elements from the invoices. These 3 elements are all located in a table/line items for all the invoices.
The 3 elements are:
1: Tariff number (digit)
2: Quantity (always a digit)
3: Total line amount (monetary value)
Please refer to below screenshot, where I have marked these field on a sample invoice.
I started this project with a template approach, based on regular expressions. This, however, was not scaleable at all and I ended up with tons of different rules.
I am hoping that machine learning can help me here - or maybe a hybrid solution?
The common denominator
In all of my invoices, despite of the different layouts, each line item will always consist of one tariff number. This tariff number is always 8 digits, and is always formatted in one the ways like below:
xxxxxxxx
xxxx.xxxx
xx.xx.xx.xx
(Where "x" is a digit from 0 - 9).
Further, as you can see on the invoice there is both a Unit Price and a Total Amount per line. The amount I will need is always the highest for each line.
The output
For each invoice like the one above, I need the output for each line. This could for example be something like this:
{
"line":"0",
"tariff":"85444290",
"quantity":"3",
"amount":"258.93"
},
{
"line":"1",
"tariff":"85444290",
"quantity":"4",
"amount":"548.32"
},
{
"line":"2",
"tariff":"76109090",
"quantity":"5",
"amount":"412.30"
}
Where to go from here?
I am not sure of what I am looking to do falls under machine learning and if so, under which category. Is it computer vision? NLP? Named Entity Recognition?
My initial thought was to:
Convert the invoice to text. (The invoices are all in textable PDFs, so I can use something like pdftotext to get the exact textual values)
Create custom named entities for quantity, tariff and amount
Export the found entities.
However, I feel like I might be missing something.
Can anyone assist me in the right direction?
Edit:
Please see below for a few more examples of how an invoice table section can look like:
Sample invoice #2
Sample invoice #3
Edit 2:
Please see below for the three sample images, without the borders/bounding boxes:
Image 1:
Image 2:
Image 3:
Here's an attempt using OpenCV, the idea is:
Obtain binary image. We load the image, enlarge using
imutils.resize to help obtain better OCR results (see Tesseract improve quality), convert to grayscale, then Otsu's threshold to obtain a binary image (1-channel).
Remove table grid lines. We create a horizontal and vertical kernels then perform morphological operations to combine adjacent text contours into a single contour. The idea is to extract a ROI row as one piece to OCR.
Extract row ROIs. We find contours then sort from top-to-bottom using imutils.contours.sort_contours. This ensures that we iterate through each row in the correct order. From here we iterate through the contours, extract the row ROI using Numpy slicing, OCR using Pytesseract, then parse the data.
Here's the visualization of each step:
Input image
Binary image
Morph close
Visualization of iterating through each row
Extracted row ROIs
Output invoice data result:
{'line': '0', 'tariff': '85444290', 'quantity': '3', 'amount': '258.93'}
{'line': '1', 'tariff': '85444290', 'quantity': '4', 'amount': '548.32'}
{'line': '2', 'tariff': '76109090', 'quantity': '5', 'amount': '412.30'}
Unfortunately, I get mixed results when trying on the 2nd and 3rd image. This method does not produce great results on the other images since the layout of the invoices are all different. However, this approach shows that it's possible to use traditional image processing techniques to extract the invoice information with the assumption that you have a fixed invoice layout.
Code
import cv2
import numpy as np
import pytesseract
from imutils import contours
import imutils
pytesseract.pytesseract.tesseract_cmd = r"C:\Program Files\Tesseract-OCR\tesseract.exe"
# Load image, enlarge, convert to grayscale, Otsu's threshold
image = cv2.imread('1.png')
image = imutils.resize(image, width=1000)
height, width = image.shape[:2]
gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
thresh = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)[1]
# Remove horizontal lines
horizontal_kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (50,1))
detect_horizontal = cv2.morphologyEx(thresh, cv2.MORPH_OPEN, horizontal_kernel, iterations=2)
cnts = cv2.findContours(detect_horizontal, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
cnts = cnts[0] if len(cnts) == 2 else cnts[1]
for c in cnts:
cv2.drawContours(thresh, [c], -1, 0, -1)
# Remove vertical lines
vertical_kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (1,50))
detect_vertical = cv2.morphologyEx(thresh, cv2.MORPH_OPEN, vertical_kernel, iterations=2)
cnts = cv2.findContours(detect_vertical, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
cnts = cnts[0] if len(cnts) == 2 else cnts[1]
for c in cnts:
cv2.drawContours(thresh, [c], -1, 0, -1)
# Morph close to combine adjacent contours into a single contour
invoice_data = []
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (85,5))
close = cv2.morphologyEx(thresh, cv2.MORPH_CLOSE, kernel, iterations=3)
# Find contours, sort from top-to-bottom
# Iterate through contours, extract row ROI, OCR, and parse data
cnts = cv2.findContours(close, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
cnts = cnts[0] if len(cnts) == 2 else cnts[1]
(cnts, _) = contours.sort_contours(cnts, method="top-to-bottom")
row = 0
for c in cnts:
x,y,w,h = cv2.boundingRect(c)
ROI = image[y:y+h, 0:width]
ROI = cv2.GaussianBlur(ROI, (3,3), 0)
data = pytesseract.image_to_string(ROI, lang='eng', config='--psm 6')
parsed = [word.lower() for word in data.split()]
if 'tariff' in parsed or 'number' in parsed:
row_data = {}
row_data['line'] = str(row)
row_data['tariff'] = parsed[-1]
row_data['quantity'] = parsed[2]
row_data['amount'] = str(max(parsed[10], parsed[11]))
row += 1
print(row_data)
invoice_data.append(row_data)
# Visualize row extraction
'''
mask = np.zeros(image.shape, dtype=np.uint8)
cv2.rectangle(mask, (0, y), (width, y + h), (255,255,255), -1)
display_row = cv2.bitwise_and(image, mask)
cv2.imshow('ROI', ROI)
cv2.imshow('display_row', display_row)
cv2.waitKey(1000)
'''
print(invoice_data)
cv2.imshow('thresh', thresh)
cv2.imshow('close', close)
cv2.waitKey()
I'm working on a similar problem in the logistics industry and trust me when I say these document tables come in myriad layouts. Numerous companies that have somewhat solved and are improving on this problem are mentioned as under
Leaders: ABBYY, AntWorks, Kofax, and WorkFusion
Major Contenders: Automation Anywhere, Celaton, Datamatics, EdgeVerve, Extract Systems, Hyland, Hyperscience, Infrrd, and Parascript
Aspirants: Ikarus, Rossum, Shipmnts(Alex), Amazon(Textract), Docsumo, Docparser, Aidock
The category I would like to put this problem under would be multi-modal learning, because both textual and image modalities contribute a good deal in this problem. Though OCR tokens play a vital role in attribute-value classification, their position on the page, spacing and inter-character distances hold as very important features in detecting table, row and column boundaries. The problem gets all the more interesting when rows break across pages, or some columns carry non-empty values.
While the academic world and conferences uses the term Intelligent Document Processing, in general for extracting both singular fields and tabular data. The former is more known by attribute-value classification and the latter is famous by table extraction or repeated-structure extraction, in research literature.
In our foray in processing these semi-structured documents over the 3 years, I feel that achieving both accuracy and scalability is a long and arduous journey. The solutions that offer scalability / 'template free' approach do have annotated corpus of semi-structured business documents in the order of tens of thousand, if not millions. Though this approach is a scalable solution, it's as good as the documents it has been trained on. If your documents hail from the logistics or insurance sector, which are known for their complex layouts, and need to be super-accurate owing to the compliance procedures, a 'template-based' solution would be the panacea to your ills. It is guaranteed to give more accuracy.
If you need links to existing research, do mention in the comments below and I'ld be happy to share them.
Also, I would recommend using pdfparser1 over pdf2text or pdfminer because the former gives character level information in digital files at significantly better performance.
Would be happy to incorporate any feedback, as this is my first answer here.
I am messing around with opencv2 for neural style transfer... In cv2.imshow("Output", output), I am able to say my picture. But when I write output to file with cv2.imwrite("my_file.jpg", output). Is it because my file extension is wrong? When I do like cv2.imwrite("my_file.jpg", input) though, it does show my original input picture. Any ideas? Thank you in advance.
# import the necessary packages
from __future__ import print_function
import argparse
import time
import cv2
import imutils
import numpy as np
from imutils.video import VideoStream
# construct the argument parser and parse the arguments
ap = argparse.ArgumentParser()
ap.add_argument("-m", "--model", required=True,
help="neural style transfer model")
ap.add_argument("-i", "--image", required=True,
help="input image to apply neural style transfer to")
args = vars(ap.parse_args())
# load the neural style transfer model from disk
print("[INFO] loading style transfer model")
net = cv2.dnn.readNetFromTorch(args["model"])
# load the input image, resize it to have a width of 600 pixels, and
# then grab the image dimensions
image = cv2.imread(args["image"])
image = imutils.resize(image, width=600)
(h, w) = image.shape[:2]
# construct a blob from the image, set the input, and then perform a
# forward pass of the network
blob = cv2.dnn.blobFromImage(image, 1.0, (w, h),
(103.939, 116.779, 123.680), swapRB=False, crop=False)
net.setInput(blob)
start = time.time()
output = net.forward()
end = time.time()
# reshape the output tensor, add back in the mean subtraction, and
# then swap the channel ordering
output = output.reshape((3, output.shape[2], output.shape[3]))
output[0] += 103.939
output[1] += 116.779
output[2] += 123.680
output /= 255.0
output = output.transpose(1, 2, 0)
# show information on how long inference took
print("[INFO] neural style transfer took {:.4f} seconds".format(
end - start))
# show the images
cv2.imshow("Input", image)
cv2.imshow("Output", output)
cv2.waitKey(0)
cv2.imwrite("dogey.jpg", output)
Only the last 4 lines of code have to deal with imshow and imwrite, all lines before are trying to modify the output picture.
The variable output represents a colored image that is composed of pixels. Each pixel is determined by three values (RGB). Depending on the representation of the image each value is chosen from the discrete range [0, 255] or continuous range [0, 1] either. However, in the following line of code, you are scaling the entries of output from the discrete range [0,255] to the "continuous" range [0,1].
output /= 255.0
While the function cv2.imshow(...) can handle images stored with float values in the range [0, 1] the cv2.imwrite(...) function cannot. You have to pass an image composed of values in the range [0, 255]. In your case, you are passing values that are all close to zero and "far" away from 255. Hence, the image is assumed as colorless and therefore black. A quick fix might be:
cv2.imwrite("dogey.jpg", 255*output)
This code runs as expected, and gives the expected output
import multiprocessing
import cv2
import os
path = r"/home/pi/Desktop/calibration.jpg"
image = cv2.imread(path)
def cvtcolor(img):
print "converting to gray ..."
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
print "converted to gray"
if True:
p = multiprocessing.Process(name='test',
target=cvtcolor,
kwargs={'img':image}
)
p.start()
p2 = multiprocessing.Process(name='test',
target=cvtcolor,
kwargs={'img':image}
)
p2.start()
outputs:
converting to gray ...
converting to gray ...
converted to gray
converted to gray
However, this code hangs when executed
import multiprocessing
import cv2
import os
path = r"/home/pi/Desktop/calibration.jpg"
image = cv2.imread(path)
def cvtcolor(img):
print "converting to gray ..."
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
print "converted to gray"
cvtcolor(image)
if True:
p = multiprocessing.Process(name='test',
target=cvtcolor,
kwargs={'img':image}
)
p.start()
the function executed in the main process proceeds, but the function executed in the "test" process hangs forever
converting to gray ...
converted to gray
converting to gray ...
I am using OpenCV version 3.2.0, installed as detailed here on Raspbian Jessie (raspberry pi)
Does anyone has an explanation / solution for this?
have a look at what is returned. If you try the BGR2GRAY directly, you will get an array with shape attribute same as input image but with only 1 color, e.g. gray. When you run the same function using multiprocessing you do not get an array returned. It will have no shape attribute, try printing the output to see what form it is in, then maybe reconstruct an image from this.
I'm using Tesseract to recognize numbers from images of a screen taken with a phone camera. I've done some preprocessing of the image: processed image, and using Tesseract, I'm able to get some mixed results. Using the following code on the above images, I get the following output: "EOE". However, with this image, processed image, I get an exact match: "39:45.8"
import cv2
import pytesseract
from PIL import Image, ImageEnhance
from matplotlib import pyplot as plt
orig_name = "time3.jpg";
image_name = "time3_.jpg";
img = cv2.imread(orig_name, 0)
img = cv2.medianBlur(img, 5)
img_th = cv2.adaptiveThreshold(img, 255,\
cv2.ADAPTIVE_THRESH_MEAN_C,cv2.THRESH_BINARY, 11, 2)
cv2.imshow('image', img_th)
cv2.waitKey(0)
cv2.imwrite(image_name, img_th)
im = Image.open(image_name)
time = pytesseract.image_to_string(im, config = "-psm 7")
print(time)
Is there anything I can do to get more consistent results?
I did three additional things to get it correct for the first Image.
You can set a whitelist for Tesseract. In your case we know that
there will only charachters from this List 01234567890.:. This
improves the accuracy significantly.
I resized the image to make it easier for tesseract.
I switched from psm mode 7 to 11 (Recoginze as much as possible)
Code:
import cv2
import pytesseract
from PIL import Image, ImageEnhance
orig_name = "./time1.jpg";
img = cv2.imread(orig_name)
height, width, channels = img.shape
imgResized = cv2.resize(img, ( width*3, height*3))
cv2.imshow("img",imgResized)
cv2.waitKey()
im = Image.fromarray(imgResized)
time = pytesseract.image_to_string(im, config ='--tessdata-dir "/home/rvq/github/tesseract/tessdata/" -c tessedit_char_whitelist=01234567890.: -psm 11 -oem 0')
print(time)
Note:
You can use Image.fromarray(imgResized) to convert an opencv image to a PIL Image. You don't have to write to disk and read it again.