Commit cb1ffc62 authored by francois's avatar francois

erp5_receipt_recognition Update bt5 following merge request advices

parent e9b63938
...@@ -77,7 +77,7 @@ ...@@ -77,7 +77,7 @@
<dictionary> <dictionary>
<item> <item>
<key> <string>text</string> </key> <key> <string>text</string> </key>
<value> <string>string:${object_url}/ReceiptConversion_convertImage</string> </value> <value> <string>string:${object_url}/ReceiptRecognition_convertImage</string> </value>
</item> </item>
</dictionary> </dictionary>
</pickle> </pickle>
......
...@@ -77,7 +77,7 @@ ...@@ -77,7 +77,7 @@
<dictionary> <dictionary>
<item> <item>
<key> <string>text</string> </key> <key> <string>text</string> </key>
<value> <string>string:${object_url}/Receipt_view</string> </value> <value> <string>string:${object_url}/ReceiptRecognition_view</string> </value>
</item> </item>
</dictionary> </dictionary>
</pickle> </pickle>
......
...@@ -6,6 +6,8 @@ to work inside erp5 and adapt to receipt binaries and with more ...@@ -6,6 +6,8 @@ to work inside erp5 and adapt to receipt binaries and with more
explanation explanation
https://github.com/tmbdev/ocropy https://github.com/tmbdev/ocropy
""" """
# pylint: disable=unpacking-non-sequence
# Pylint is confused by ocropy.
import numpy as np import numpy as np
import scipy.ndimage as ndi import scipy.ndimage as ndi
...@@ -14,9 +16,10 @@ from matplotlib import pylab ...@@ -14,9 +16,10 @@ from matplotlib import pylab
import matplotlib.image as mpimg import matplotlib.image as mpimg
import scipy.stats as stats import scipy.stats as stats
import re import re
import cPickle
import ocrolib import ocrolib
def getReceiptValue(self, image_data): def getReceiptValue(self, image_data, model_name = "en-default.pyrnn"):
""" """
Function called from an erp5 script through externalMethod Function called from an erp5 script through externalMethod
that take an image and its name and save its binarized that take an image and its name and save its binarized
...@@ -27,20 +30,46 @@ def getReceiptValue(self, image_data): ...@@ -27,20 +30,46 @@ def getReceiptValue(self, image_data):
Represent the erp5 object from which externalmethods or module Represent the erp5 object from which externalmethods or module
objects can be called objects can be called
- image_data: - image_data:
base64 representation of the image to analyse Representation of the image to analyse
@return: @return:
- ret: float - anon: float
Represent total value paid on the receipt Represent total value paid on the receipt
---------------------------- ----------------------------
This function return the total value of the receipt in euros. This function look for euros only and return a price with a two digit
precison like "135.79" or "43,89".
""" """
image_as_string = StringIO.StringIO(image_data) image_as_string = StringIO.StringIO(image_data)
image_as_array = mpimg.imread(image_as_string, format = 'JPG') image_as_array = mpimg.imread(image_as_string, format = 'JPG')
line_list, cleared = getLinesFromPicture(image_as_array) line_list, cleared = getLinesFromPicture(image_as_array)
# Start the neural network # Start the neural network
network, lnorm = initRnnModel() network, lnorm = initRnnModel(model_name)
return findReceiptValue(line_list, cleared, network, lnorm)
def findReceiptValue(line_list, cleared, network, lnorm):
"""
Function that run the neural network through the receipt and extract
meaningfull value
-----------------------------
@args:
- lines: array list
Represent lines of text that will be extracted
from the image
- cleared:2D array
Represent binarized image cropped and cleaned,
from which we will extract text lines
- network: lstm object
Represent the trained neural net
- lnorm: method from lstm object
Represent the size of the lstm object. Is used to scale the objects
to recognize from original size to the average network object.
@return:
- anon: float
Represent total value paid on the receipt
-----------------------------
This function can bemodified to add more field to detect. It might be
possible to run a classification neural net on the result.
"""
value_list = [] value_list = []
tofind = r"(EUR)|€|(TOT)" tofind = r"(EUR)|€|(TOT)"
for _, line in enumerate(line_list): for _, line in enumerate(line_list):
...@@ -54,9 +83,28 @@ def getReceiptValue(self, image_data): ...@@ -54,9 +83,28 @@ def getReceiptValue(self, image_data):
value_list += [float(char.replace(',', '.')) for char in number] value_list += [float(char.replace(',', '.')) for char in number]
except ValueError: except ValueError:
pass pass
return round(max(value_list), 2) return round(max(value_list), 2)
def getRnnModelFromDataStream(self, model_name="en-default.pyrnn"):
"""
This function load a neural network from a dataStream
----------------------------
@args:
- model_name: string, default: en-default.pyrnn
Id of the object in data_stream_module that contain the rnn model
@return:
- network: lstm object
Represent the trained neural net
- lnorm: method from lstm object
Represent the size of the lstm object. Is used to scale the objects
to recognize from original size to the average network object.
----------------------------
WARNING: This function present a security issue and should NOT be called with
an user-defined model name (see cpickle security issue)
"""
network = cPickle.loads(self.data_stream_module[model_name].getData())
lnorm = getattr(network, "lnorm", None)
return network, lnorm
def initRnnModel(model_name = "en-default.pyrnn"): def initRnnModel(model_name = "en-default.pyrnn"):
""" """
...@@ -65,7 +113,7 @@ def initRnnModel(model_name = "en-default.pyrnn"): ...@@ -65,7 +113,7 @@ def initRnnModel(model_name = "en-default.pyrnn"):
---------------------------- ----------------------------
@args: @args:
- model_name: string, default: en-default.pyrnn - model_name: string, default: en-default.pyrnn
Id of the object in data_stream_module that contain the rnn model Id of the object in the filesystem that contain the rnn model
@return: @return:
- network: lstm object - network: lstm object
Represent the trained neural net Represent the trained neural net
...@@ -108,8 +156,8 @@ def getLinesFromPicture(image_as_array): ...@@ -108,8 +156,8 @@ def getLinesFromPicture(image_as_array):
independant picture independant picture
""" """
grey_image = convertGreyscale(image_as_array) grey_image = convertGreyscale(image_as_array)
flattened_image = imageTransformation(grey_image) cropped_image = cropImage(grey_image)
binarized_image = imageBinarization(flattened_image) binarized_image = imageBinarization(cropped_image)
binary = 1 - binarized_image binary = 1 - binarized_image
cleaned, scale = removeObjects(binary) cleaned, scale = removeObjects(binary)
angle = getEstimatedSkewAngle(cleaned, np.linspace(-4, 4, 24)) angle = getEstimatedSkewAngle(cleaned, np.linspace(-4, 4, 24))
...@@ -293,26 +341,6 @@ def removeObjects(binarized): ...@@ -293,26 +341,6 @@ def removeObjects(binarized):
binarized = np.minimum(binarized, 1 - (sums > 0) * (sums < scale)) binarized = np.minimum(binarized, 1 - (sums > 0) * (sums < scale))
return binarized, scale return binarized, scale
def getImageWhitelevel(image):
"""
Function that help flatten the image by estimating locals
whitelevels. This remove local extremes and give an image with
homogenous background and no details
------------------------------
@args:
- image: 2D array
Represent a greyscale image
@return:
- white_image: 2D array
Represent a greyscale image with no local extreme
------------------------------
This function result will be substracted from the original image
to make that only local extremes stand out.
"""
white_image = ndi.filters.percentile_filter(image, 50, size = (80, 2))
white_image = ndi.filters.percentile_filter(white_image, 50, size = (2, 80))
return white_image
def getEstimatedSkewAngle(image, angle_list): def getEstimatedSkewAngle(image, angle_list):
""" """
Function that estimate at which angle the image is the most Function that estimate at which angle the image is the most
...@@ -343,8 +371,37 @@ def getEstimatedSkewAngle(image, angle_list): ...@@ -343,8 +371,37 @@ def getEstimatedSkewAngle(image, angle_list):
_, angle = max(estimates) _, angle = max(estimates)
return angle return angle
def removeBackground(image, percentile=50):
"""
Function that help flatten the image by estimating locals
whitelevels. This remove local extremes and give an image with
homogenous background and no details
------------------------------
@args:
- image: 2D array
Represent a greyscale image
- percentile: integer between -100 and 100
A percentile filter with a value of 50 is basically a
median filter, value of 0 is a minimum filter and with
a value of 100 a maximum filter
@return:
- 2D array
Represent a greyscale image with no local extreme
------------------------------
The filter result will be substracted from the original image
to make that only local extremes stand out.
A Kuwahara filter might give better results.
"""
# Reduce extreme differences in the greyscale image
image = image - pylab.amin(image)
image /= pylab.amax(image)
white_image = ndi.filters.percentile_filter(image, percentile, size=(80, 2))
white_image = ndi.filters.percentile_filter(white_image, percentile, size=(2, 80))
# Get the difference between the whiteleveled image and the
# original one and put them betewwn 0 an 1
return np.clip(image - white_image + 1, 0, 1)
def imageTransformation(grey): def cropImage(image):
""" """
Function that perform cropping and flattening -- Removing Function that perform cropping and flattening -- Removing
homogenous background and small extremes-- on an image. homogenous background and small extremes-- on an image.
...@@ -360,22 +417,17 @@ def imageTransformation(grey): ...@@ -360,22 +417,17 @@ def imageTransformation(grey):
homogenous background homogenous background
""" """
# Reduce extreme differences in the greyscale image # Reduce extreme differences in the greyscale image
image = grey - pylab.amin(grey) white_image = removeBackground(image)
image /= pylab.amax(image)
white_image = getImageWhitelevel(image)
# Get the difference between the whiteleveled image and the
# original one and put them betewwn 0 an 1
flat = np.clip(image - white_image + 1, 0, 1)
# Calculate coordinate to crop the image, can be done in another # Calculate coordinate to crop the image, can be done in another
# function to improve readability # function to improve readability
mask = ndi.gaussian_filter( mask = ndi.gaussian_filter(
flat, 7.0) < 0.9 * np.amax(flat) white_image, 7.0) < 0.9 * np.amax(white_image)
coords = np.argwhere(mask) coords = np.argwhere(mask)
# Bounding box of kept pixels. # Bounding box of kept pixels.
x_min, y_min = coords.min(axis = 0) x_min, y_min = coords.min(axis = 0)
x_max, y_max = coords.max(axis = 0) x_max, y_max = coords.max(axis = 0)
return flat[x_min - 10 : x_max + 10, y_min - 10 : y_max + 10] return white_image[x_min - 10 : x_max + 10, y_min - 10 : y_max + 10]
def imageBinarization(flattened_image): def imageBinarization(flattened_image):
......
...@@ -45,10 +45,7 @@ ...@@ -45,10 +45,7 @@
<item> <item>
<key> <string>text_content_warning_message</string> </key> <key> <string>text_content_warning_message</string> </key>
<value> <value>
<tuple> <tuple/>
<string>W:243, 2: Attempting to unpack a non-sequence defined at line 181 of scipy.ndimage.measurements (unpacking-non-sequence)</string>
<string>W:272, 2: Attempting to unpack a non-sequence defined at line 181 of scipy.ndimage.measurements (unpacking-non-sequence)</string>
</tuple>
</value> </value>
</item> </item>
<item> <item>
......
<allowed_content_type_list> <allowed_content_type_list>
<portal_type id="Receipt Recognition Module"> <portal_type id="Receipt Recognition Module">
<item>Receipt</item> <item>Receipt Recognition</item>
</portal_type> </portal_type>
</allowed_content_type_list> </allowed_content_type_list>
\ No newline at end of file
<property_sheet_list> <property_sheet_list>
<portal_type id="Receipt"> <portal_type id="Receipt Recognition">
<item>Document</item> <item>Document</item>
</portal_type> </portal_type>
</property_sheet_list> </property_sheet_list>
\ No newline at end of file
...@@ -28,7 +28,7 @@ ...@@ -28,7 +28,7 @@
</item> </item>
<item> <item>
<key> <string>id</string> </key> <key> <string>id</string> </key>
<value> <string>Receipt</string> </value> <value> <string>Receipt Recognition</string> </value>
</item> </item>
<item> <item>
<key> <string>init_script</string> </key> <key> <string>init_script</string> </key>
......
...@@ -2,17 +2,17 @@ image = context.getFollowUpValue() ...@@ -2,17 +2,17 @@ image = context.getFollowUpValue()
if image is not None: if image is not None:
try: try:
total = container.ReceiptRecognition_getReceiptValue(image.getData()) total = container.ReceiptRecognition_getReceiptValue(image.getData())
msg = "Total found" message = "Total found"
context.edit( context.edit(
total = total, total = total,
) )
except ValueError as e: except ValueError as e:
msg = "Could not find value, please submit it manually" message = "Could not find value, please submit it manually"
else: else:
msg = "Cannot find the image" message = "Cannot find the image"
if batch_mode: if batch_mode:
return return
context.Base_redirect( context.Base_redirect(
'view', keep_items = dict(portal_status_message=msg, my_source="test")) 'view', keep_items = dict(portal_status_message=message, my_source="test"))
...@@ -54,7 +54,7 @@ ...@@ -54,7 +54,7 @@
</item> </item>
<item> <item>
<key> <string>id</string> </key> <key> <string>id</string> </key>
<value> <string>ReceiptConversion_convertImage</string> </value> <value> <string>ReceiptRecognition_convertImage</string> </value>
</item> </item>
</dictionary> </dictionary>
</pickle> </pickle>
......
...@@ -89,7 +89,7 @@ ...@@ -89,7 +89,7 @@
</item> </item>
<item> <item>
<key> <string>id</string> </key> <key> <string>id</string> </key>
<value> <string>Receipt_view</string> </value> <value> <string>ReceiptRecognition_view</string> </value>
</item> </item>
<item> <item>
<key> <string>method</string> </key> <key> <string>method</string> </key>
......
Receipt Recognition Module | view Receipt Recognition Module | view
Receipt | receipt_convert Receipt Recognition | receipt_convert
Receipt | view Receipt Recognition | view
\ No newline at end of file \ No newline at end of file
Receipt Recognition Module | Receipt Receipt Recognition Module | Receipt Recognition
\ No newline at end of file \ No newline at end of file
Receipt Receipt Recognition
Receipt Recognition Module Receipt Recognition Module
\ No newline at end of file
Receipt | Document Receipt Recognition | Document
\ No newline at end of file \ No newline at end of file
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment