Hi friends following is the code for Part 19 of the PyQt5 learning series. For details visit the pyshine channel below.
How to improve accuracy of tesseract
The default Page segmentation method (psm) in tesseract is page of text. It means that tesseract expects a page of text when it segments an image. If you’re going to crop small region for OCR, try a different segmentation mode, using the –psm argument. Note that adding a white border to text which is too tightly cropped may also help. We have see a list of available psm modes using tesseract --help-psm.
(base) PS C:\Users\ps\Desktop> tesseract --help-psm
Page segmentation modes:
0 Orientation and script detection (OSD) only.
1 Automatic page segmentation with OSD.
2 Automatic page segmentation, but no OSD, or OCR. (not implemented)
3 Fully automatic page segmentation, but no OSD. (Default)
4 Assume a single column of text of variable sizes.
5 Assume a single uniform block of vertically aligned text.
6 Assume a single uniform block of text.
7 Treat the image as a single text line.
8 Treat the image as a single word.
9 Treat the image as a single word in a circle.
10 Treat the image as a single character.
11 Sparse text. Find as much text as possible in no particular order.
12 Sparse text with OSD.
13 Raw line. Treat the image as a single text line,
bypassing hacks that are Tesseract-specific.
There are a number of other factors that can help in increasing the accuracy of tesseract
- Rotation / Deskewing - rotating image to align the text horizontally and deskewing to make better orientation of characters
- Rescaling - resizing the image
- Binarisation - convert an image to black and white
- Dilation and Erosion - fixing image artifacts via image processing
- Borders - removing dark borders
- Transparency / Alpha channel - removing the alpha channel from image before ocr
- Noise Removal - removing noise like salt and pepper etc. before binarization
- Inverting images - using dark text on lighter background helps in improving accuracy
- oem - OCR Engine Mode
The available oem engines can be found via tesseract --help-oem:
(py38) PS C:\Users\ps\Desktop\ocr-19> tesseract --help-oem
OCR Engine modes:
0 Legacy engine only.
1 Neural nets LSTM engine only.
2 Legacy + LSTM engines.
3 Default, based on what is available.
Tesseract 4 contains LSTM neural net mode which mostly works best, but you are free to try. It is highly recommended to read this document for more details.
In the following gui.py code we will use following configurations custom_oem_psm_config = r'--oem 3 --psm 6' and then use like this pytesseract.image_to_string(image, config=custom_oem_psm_config)
main.ui
<?xml version="1.0" encoding="UTF-8"?>
<ui version="4.0">
<class>MainWindow</class>
<widget class="QMainWindow" name="MainWindow">
<property name="geometry">
<rect>
<x>0</x>
<y>0</y>
<width>1536</width>
<height>1006</height>
</rect>
</property>
<property name="windowTitle">
<string>PyShine OCR GUI</string>
</property>
<widget class="QWidget" name="centralwidget">
<layout class="QGridLayout" name="gridLayout_3">
<item row="0" column="0">
<widget class="QScrollArea" name="scrollArea">
<property name="verticalScrollBarPolicy">
<enum>Qt::ScrollBarAlwaysOn</enum>
</property>
<property name="horizontalScrollBarPolicy">
<enum>Qt::ScrollBarAlwaysOn</enum>
</property>
<property name="widgetResizable">
<bool>true</bool>
</property>
<widget class="QWidget" name="scrollAreaWidgetContents">
<property name="geometry">
<rect>
<x>0</x>
<y>0</y>
<width>1464</width>
<height>560</height>
</rect>
</property>
<layout class="QGridLayout" name="gridLayout">
<item row="0" column="0">
<widget class="QLabel" name="label_2">
<property name="styleSheet">
<string notr="true">background-color: rgb(170, 255, 255);</string>
</property>
<property name="text">
<string>Photo</string>
</property>
<property name="alignment">
<set>Qt::AlignCenter</set>
</property>
</widget>
</item>
</layout>
</widget>
</widget>
</item>
</layout>
</widget>
<widget class="QStatusBar" name="statusbar"/>
<widget class="QDockWidget" name="dockWidget">
<attribute name="dockWidgetArea">
<number>8</number>
</attribute>
<widget class="QWidget" name="dockWidgetContents_2">
<layout class="QGridLayout" name="gridLayout_2">
<item row="1" column="0">
<widget class="QTextEdit" name="textEdit"/>
</item>
<item row="0" column="0">
<layout class="QHBoxLayout" name="horizontalLayout">
<item>
<widget class="QLabel" name="label_3">
<property name="text">
<string>Font Size:</string>
</property>
</widget>
</item>
<item>
<widget class="QComboBox" name="comboBox_2"/>
</item>
<item>
<spacer name="horizontalSpacer">
<property name="orientation">
<enum>Qt::Horizontal</enum>
</property>
<property name="sizeHint" stdset="0">
<size>
<width>40</width>
<height>20</height>
</size>
</property>
</spacer>
</item>
<item>
<widget class="QPushButton" name="pushButton_2">
<property name="text">
<string>Zoom In</string>
</property>
</widget>
</item>
<item>
<widget class="QPushButton" name="pushButton_3">
<property name="text">
<string>Zoom Out</string>
</property>
</widget>
</item>
<item>
<widget class="QPushButton" name="pushButton_4">
<property name="text">
<string>Reset</string>
</property>
</widget>
</item>
<item>
<widget class="QLabel" name="label">
<property name="text">
<string>Select Language:</string>
</property>
</widget>
</item>
<item>
<widget class="QComboBox" name="comboBox"/>
</item>
<item>
<widget class="QPushButton" name="pushButton">
<property name="text">
<string>Open Image</string>
</property>
</widget>
</item>
</layout>
</item>
</layout>
</widget>
</widget>
</widget>
<resources/>
<connections/>
</ui>
gui.py
# Welcome to PyShine
import pytesseract
import cv2, os,sys
from PIL import Image
import PyQt5
from PyQt5.QtGui import *
from PyQt5.QtCore import *
from PyQt5.QtWidgets import *
from PyQt5 import uic
from PyQt5 import QtCore,QtGui,QtWidgets
import glob
from PIL import ImageQt
import numpy as np
from tesserocr import PyTessBaseAPI
# Here we will get the path of the tessdata
# For 64 bit installation of tesseract OCR
language_path = 'C:\\Program Files\\Tesseract-OCR\\tessdata\\'
language_path_list = glob.glob(language_path+"*.traineddata")
custom_oem_psm_config = r'--oem 3 --psm 6' # Assume a single uniform block of text.
language_names_list = []
for path in language_path_list:
base_name = os.path.basename(path)
base_name = os.path.splitext(base_name)[0]
language_names_list.append(base_name)
languages = ['afr', 'amh', 'ara', 'asm', 'aze', 'aze_cyrl', 'bel', 'ben', 'bod',
'bos', 'bre', 'bul', 'cat', 'ceb', 'ces', 'chi_sim', 'chi_sim_vert',
'chi_tra', 'chi_tra_vert', 'chr', 'cym', 'dan', 'ell',
'eng', 'enm', 'epo', 'equ', 'est', 'eus', 'fas', 'fil', 'fin', 'fra',
'frk', 'frm', 'fry', 'gla', 'gle', 'glg', 'grc', 'guj', 'hat', 'heb',
'hin', 'hrv', 'hun', 'hye', 'iku', 'ind', 'isl', 'ita', 'ita_old',
'jav', 'jpn', 'jpn_vert', 'kan', 'kat', 'kat_old', 'kaz', 'khm',
'kmr', 'kor', 'lao', 'lav', 'lit', 'ltz', 'mal', 'mar', 'mkd',
'mlt', 'mon', 'mri', 'msa', 'mya', 'nep', 'nld', 'nor', 'ori',
'osd', 'pan', 'pol', 'por', 'pus', 'que', 'ron', 'rus', 'san',
'sin', 'slk', 'slv', 'snd', 'spa', 'spa_old', 'sqi', 'srp',
'srp_latn', 'sun', 'swa', 'swe', 'syr', 'tam', 'tat', 'tel',
'tgk', 'tha', 'tir', 'ton', 'tur', 'uig', 'ukr', 'urd', 'uzb', 'uzb_cyrl', 'vie', 'yid', 'yor']
def get_score(image, language):
score =0
with PyTessBaseAPI() as api:
api.Init(lang=language)
api.SetImageFile(image)
awc = list(api.AllWordConfidences())
score = sum(awc)/(1e-20+float(len(awc)))
return score
font_list = []
font = 2
for font in range(110):
font+=2
font_list.append(str(font))
# print('Font list:',font_list)
class PyShine_OCR_APP(QtWidgets.QMainWindow):
def __init__(self):
QtWidgets.QMainWindow.__init__(self)
self.ui = uic.loadUi('main.ui',self)
self.image = None
self.ui.pushButton.clicked.connect(self.open)
self.ui.pushButton_2.clicked.connect(self.zoom_in)
self.ui.pushButton_3.clicked.connect(self.zoom_out)
self.ui.pushButton_4.clicked.connect(self.reset_zoom)
self.scrollArea = QtWidgets.QScrollArea()
self.scrollArea.setWidget(self.ui.label_2)
self.scrollArea.setVisible(True)
self.ui.setCentralWidget(self.scrollArea)
self.rubberBand = QRubberBand(QRubberBand.Rectangle,self)
self.ui.label_2.setMouseTracking(True)
self.ui.label_2.setScaledContents(True)
self.ui.label_2.installEventFilter(self)
self.ui.label_2.setAlignment(PyQt5.QtCore.Qt.AlignTop)
self.language = 'eng'
self.comboBox.addItems(languages)
self.comboBox.currentIndexChanged['QString'].connect(self.update_now)
self.comboBox.setCurrentIndex(languages.index(self.language))
self.font_size = '20'
self.text = ''
self.comboBox_2.addItems(font_list)
self.comboBox_2.currentIndexChanged['QString'].connect(self.update_font_size)
self.comboBox_2.setCurrentIndex(font_list.index(self.font_size))
self.ui.textEdit.setFontPointSize(int(self.font_size))
self.setAcceptDrops(True)
self.zoom_in,self.zoom_out,self.reset_zoom = False,False, False
self.ui.label_2.setScaledContents(True)
self.scale_factor = 1.0
self.ui.label_2.adjustSize()
self.image = ''
def zoom_in(self):
self.zoom_in, self.zoom_out,self.reset_zoom = True, False, False
self.zoom()
def zoom_out(self):
self.zoom_in, self.zoom_out,self.reset_zoom = False, True, False
self.zoom()
def reset_zoom(self):
self.zoom_in, self.zoom_out,self.reset_zoom = False, False, True
self.zoom()
def zoom(self):
if (self.scale_factor > 3.0 and self.zoom_in) or (self.scale_factor < 0.3 and self.zoom_out):
return
try:
if self.zoom_in:
frame = cv2.cvtColor(self.image, cv2.COLOR_BGR2RGB)
image = QImage(frame,frame.shape[1],frame.shape[0],frame.strides[0],QImage.Format_RGB888)
self.ui.label_2.setPixmap(QPixmap.fromImage(image))
self.scale_factor *= 1.1
newScaleImage = self.scale_factor * self.ui.label_2.pixmap().size()
self.ui.label_2.resize(newScaleImage)
self.ui.label_2.setPixmap(QPixmap.fromImage(image).scaled(
self.ui.label_2.size(), QtCore.Qt.IgnoreAspectRatio,
QtCore.Qt.SmoothTransformation))
elif self.zoom_out:
frame = cv2.cvtColor(self.image, cv2.COLOR_BGR2RGB)
image = QImage(frame,frame.shape[1],frame.shape[0],frame.strides[0],QImage.Format_RGB888)
self.ui.label_2.setPixmap(QPixmap.fromImage(image))
self.scale_factor /=1.1
newScaleImage = self.scale_factor * self.ui.label_2.pixmap().size()
self.ui.label_2.resize(newScaleImage)
self.ui.label_2.setPixmap(QPixmap.fromImage(image).scaled(
self.ui.label_2.size(), QtCore.Qt.IgnoreAspectRatio,
QtCore.Qt.SmoothTransformation))
elif self.reset_zoom:
frame = cv2.cvtColor(self.image, cv2.COLOR_BGR2RGB)
image = QImage(frame,frame.shape[1],frame.shape[0],frame.strides[0],QImage.Format_RGB888)
self.ui.label_2.setPixmap(QPixmap.fromImage(image))
self.scale_factor = 1.0
self.ui.label_2.resize(self.ui.label_2.pixmap().size())
self.ui.label_2.setPixmap(QPixmap.fromImage(image).scaled(
self.ui.label_2.size(), QtCore.Qt.IgnoreAspectRatio,
QtCore.Qt.SmoothTransformation))
except Exception as e:
print(e)
pass
def update_now(self,value):
self.language = value
print('Language Selected as:',self.language)
def update_font_size(self,value):
self.font_size = value
self.ui.textEdit.setFontPointSize(int(self.font_size))
self.ui.textEdit.setText(str(self.text))
def open(self):
filename = QFileDialog.getOpenFileName(self,'Select File')
self.image = cv2.imread(str(filename[0]))
frame = cv2.cvtColor(self.image, cv2.COLOR_BGR2RGB)
image = QImage(frame,frame.shape[1],frame.shape[0],frame.strides[0],QImage.Format_RGB888)
self.ui.label_2.setPixmap(QPixmap.fromImage(image))
image = ImageQt.fromqpixmap(self.ui.label_2.pixmap())
image = cv2.cvtColor(np.array(image), cv2.COLOR_RGB2BGR)
self.ui.label_2.adjustSize()
def image_to_text(self,crop_cvimage):
gray = cv2.cvtColor(crop_cvimage,cv2.COLOR_BGR2GRAY)
gray = cv2.medianBlur(gray,1)
crop = Image.fromarray(gray)
output = [get_score('cropped.png',language) for language in languages] # get scores of all languages
index = np.array(output).argmax() # get index of max in the list
print(f'detected language is {languages[index]} score {max(output)}')
self.language = languages[index]
self.comboBox.setCurrentIndex(languages.index(self.language))
#text = pytesseract.image_to_string(crop,lang = self.language)
text = pytesseract.image_to_string(crop,lang = self.language, config=custom_oem_psm_config)
print('Text:',text)
return text
def eventFilter(self,source,event):
width = 0
height = 0
if (event.type() == QEvent.MouseButtonPress and source is self.ui.label_2):
self.org = self.mapFromGlobal(event.globalPos())
self.left_top = event.pos()
self.rubberBand.setGeometry(QRect(self.org,QSize()))
self.rubberBand.show()
elif (event.type() == QEvent.MouseMove and source is self.ui.label_2):
if self.rubberBand.isVisible():
self.rubberBand.setGeometry(QRect(self.org,self.mapFromGlobal(event.globalPos())).normalized())
elif(event.type() == QEvent.MouseButtonRelease and source is self.ui.label_2):
if self.rubberBand.isVisible():
self.rubberBand.hide()
rect = self.rubberBand.geometry()
self.x1 = self.left_top.x()
self.y1 = self. left_top.y()
width = rect.width()
height = rect.height()
self.x2 = self.x1+ width
self.y2 = self.y1+ height
if width >=10 and height >= 10 and self.image is not None:
image = ImageQt.fromqpixmap(self.ui.label_2.pixmap())
image = cv2.cvtColor(np.array(image), cv2.COLOR_RGB2BGR)
self.crop = image[self.y1:self.y2, self.x1:self.x2]
cv2.imwrite('cropped.png',self.crop)
self.text = self.image_to_text(self.crop)
self.ui.textEdit.setText(str(self.text))
else:
self.rubberBand.hide()
else:
return 0
return QWidget.eventFilter(self,source,event)
# www.pyshine.com
app = QtWidgets.QApplication(sys.argv)
mainWindow = PyShine_OCR_APP()
mainWindow.show()
sys.exit(app.exec_())
Useful Posts