def process_image(imag):
# %matplotlib inline
IMG = imag
plt.rcParams['figure.figsize'] = (15.0, 10.0)
"""### Global Variables"""
"""## Load image"""
image = cv2.cvtColor(cv2.imread(IMG), cv2.COLOR_BGR2RGB)
implt(image)
# Crop image and get bounding boxes
crop = page.detection(image)
implt(crop)
boxes = words.detection(crop)
lines = words.sort_words(boxes)
implt(crop)
output_file = open("templates/output.html", 'w+')
output_file.write("")
output_file.close()
output_file = open("templates/output.html", 'a+')
for line in lines:
print(" ".join(
[recognise(crop[y1:y2, x1:x2]) for (x1, y1, x2, y2) in line]))
for (x1, y1, x2, y2) in line:
text_det = recognise(crop[y1:y2, x1:x2])
output_file.write(text_det + " ")
output_file.write("\n")
output_file.close()
def recognize():
if request.method == 'POST':
# start = current_milli_time()
if 'image' not in request.files:
print('No file part')
return render_template('index.html')
file = request.files['image']
if file.filename == '':
print('No selected file')
return render_template('index.html')
if file:
file.save(os.path.join('images',file.filename))
filename = file.filename
# load the trained model
charClass = Graph(MODEL_LOC)
# load the saved image
image = cv2.cvtColor(cv2.imread("images/"+filename), cv2.COLOR_BGR2RGB)
crop = page.detection(image)
bBoxes = words.detection(crop)
cycler = Cycler.Cycler(crop,bBoxes,charClass)
allWords = []
for i in range(len(bBoxes)-1):
allWords.append(cycler.idxImage(i))
# print(allWords[i])
# stop = current_milli_time()
# print(stop - start)
return jsonify(allWords=allWords)
def apply_with_source_info(self, im, source_info):
boxes = words.detection(im)
lines = words.sort_words(boxes)
print('lines')
print(lines)
self.add_info('bbxy', boxes)
self.add_info('lines', lines)
return im
def get_bboxes(img_path):
img = get_image_from_file(img_path)
img_gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
img[:, :, 0][img_gray > 180] = 255
img[:, :, 0][img_gray <= 180] = 0
img[:, :, 1][img_gray > 180] = 255
img[:, :, 1][img_gray <= 180] = 0
img[:, :, 2][img_gray > 180] = 255
img[:, :, 2][img_gray <= 180] = 0
bboxes = words.detection(img)
return bboxes
def test_recognize():
charClass = Graph(MODEL_LOC)
# load the saved image
image = cv2.cvtColor(cv2.imread(FILE_LOC), cv2.COLOR_BGR2RGB)
crop = page.detection(image)
bBoxes = words.detection(crop)
cycler = Cycler.Cycler(crop, bBoxes, charClass)
allWords = []
for i in range(len(bBoxes) - 1):
allWords.append(cycler.idxImage(i))
isStructEmpty = structEmpty(allWords)
assert isStructEmpty == False
def main():
Infilename = sys.argv[1]
image = cv2.cvtColor(cv2.imread(Infilename), cv2.COLOR_BGR2RGB)
crop = page.detection(image)
boxes = words.detection(crop)
lines = words.sort_words(boxes)
crop = cv2.cvtColor(crop, cv2.COLOR_RGB2GRAY)
imLines = []
for line in lines:
imLine = []
for (x1, y1, x2, y2) in line:
imLine.append(crop[y1:y2, x1:x2])
imLines.append(imLine)
decoderType = DecoderType.WordBeamSearch
#decoderType = DecoderType.BeamSearch
#decoderType = DecoderType.BestPath
model = Model(open('../model/charList.txt').read(),
decoderType,
mustRestore=True)
file1 = open("myfile.txt", "w")
recognizedL = []
print(
"-------------------Predicted Handwritten Text-------------------------"
)
for line in imLines:
imgs = []
for word in line:
imgs.append(preprocess(word, Model.imgSize))
batch = Batch(None, imgs)
(recognized, probability) = model.inferBatch(batch, True)
l = ""
for pw in recognized:
l += pw
l += ' '
print(pw, end=" ")
print()
l += '\n'
recognizedL.append(l)
file1.writelines(recognizedL)
file1.close()
def preprocess(empty_img_path, img_path):
bboxes = get_bboxes(empty_img_path)
img = get_image_from_file(img_path)
img = cv2.cvtColor(img, cv2.COLOR_RGB2GRAY)
img[img > 180] = 255
img[img <= 180] = 0
for bx in bboxes:
x1, y1, x2, y2 = bx
img[0:256, 0:x2 + 5] = 255
img_rgb = np.zeros((HEIGHT, WIDTH, 3))
img_rgb[:, :, 0] = img
img_rgb[:, :, 1] = img
img_rgb[:, :, 2] = img
bboxes = words.detection(img_rgb)
xx1, yy1, xx2, yy2 = np.inf, np.inf, 0, 0
for bx in bboxes:
x1, y1, x2, y2 = bx
if x1 < xx1:
xx1 = x1
if y1 < yy1:
yy1 = y1
if x2 > xx2:
xx2 = x2
if y2 > yy2:
yy2 = y2
img = cv2.resize(img[yy1:yy2, xx1:xx2], dsize=(O_WIDTH, O_HEIGHT))
img = img.astype(np.float32) / 255
img[img > 0.75] = 1
img[img <= 0.75] = 0
return img
def visual_feedback(self, im):
plt.rcParams['figure.figsize'] = (15.0, 10.0)
# Crop image and get bounding boxes
boxes = words.detection(im)
if self.values['inverted colors feedback']:
im[:,:,:] = 255-im[:,:,:]
for box in boxes:
cv2.rectangle(im,(box[0], box[1]),(box[2], box[3]),(0,255,0),3)
'''
for line in lines:
if self.values['model'] == 'base_model':
print(" ".join([self.recognise_char_sep_model(crop[y1:y2, x1:x2]) for (x1, y1, x2, y2) in line]))
else:
print(" ".join([self.recognise_ctc(crop[y1:y2, x1:x2]) for (x1, y1, x2, y2) in line]))
'''
return im
def textRecog(infilename):
image = cv2.cvtColor(cv2.imread(infilename), cv2.COLOR_BGR2RGB)
crop = page.detection(image)
boxes = words.detection(crop)
lines = words.sort_words(boxes)
crop = cv2.cvtColor(crop, cv2.COLOR_RGB2GRAY)
imLines = []
for line in lines:
imLine = []
for (x1, y1, x2, y2) in line:
imLine.append(crop[y1:y2, x1:x2])
imLines.append(imLine)
#decoderType = DecoderType.WordBeamSearch
#decoderType = DecoderType.BeamSearch
decoderType = DecoderType.BestPath
model = Model(open('./model/charList.txt').read(),
decoderType,
mustRestore=True)
#file1 = open("myfile.txt", "w")
recognizedText = ""
for line in imLines:
imgs = []
for word in line:
imgs.append(preprocess(word, Model.imgSize))
batch = Batch(None, imgs)
(recognized, probability) = model.inferBatch(batch, True)
l = ""
for pw in recognized:
l += pw
l += ' '
recognizedText += l
return recognizedText
def fun(list1):
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import tensorflow as tf
import cv2
import random
# Import costume functions, corresponding to notebooks
from ocr.normalization import imageNorm, letterNorm
from ocr import page, words
#from ocr import charSeg
from ocr.helpers import implt, resize
from ocr.tfhelpers import Graph
from ocr.datahelpers import idx2char
from src import gsmain
import glob
import os
# ### Global Variables
for x_file in list1:
tf.reset_default_graph()
str1 = ''.join(x_file)
# Settings
IMG = str1 # 1, 2, 3
# ## Load image
image = cv2.cvtColor(cv2.imread(IMG), cv2.COLOR_BGR2RGB)
implt(image)
# Crop image and get bounding boxes
crop = page.detection(image)
implt(crop)
bBoxes = words.detection(crop)
lines = words.sort_words(bBoxes)
filelist = glob.glob("./new3/*.png")
for file in filelist:
os.remove(file)
indeximg = 0
nl = 0
for line in lines:
for (x1, y1, x2, y2) in line:
cv2.imwrite("new3/" + str(indeximg) + ".png", crop[y1:y2,
x1:x2])
#implt(cv2.imread("outcheck.png"))
indeximg = indeximg + 1
#gsmain.FilePaths.fnInfer = ["outcheck.png"]
#wordreco = gsmain.main()
#file.write(wordreco + ' ')
cv2.imwrite("new3/" + str(nl) + "space.png", crop[y1:y2, x1:x2])
nl = nl + 1
#Get all segmented words
list2 = glob.glob("./new3/*.png")
list2.sort(key=os.path.getmtime)
'''list2=list()
for ii in range(0,indeximg):
list2.append("")'''
#all files which have to be infer are loaded
rand_num = random.randint(100, 1000)
cv2.imwrite('final_output/' + str(rand_num) + '.jpg', image)
for i in range(0, len(list2)):
gsmain.FilePaths.fnInfer = gsmain.FilePaths.fnInfer + [list2[i]]
gsmain.main('final_output/' + str(rand_num))
charClass = Graph(MODEL_LOC)
# ## Load image
# In[4]:
image = cv2.cvtColor(cv2.imread("test/%s.jpg" % IMG), cv2.COLOR_BGR2RGB)
implt(image)
# In[5]:
# Crop image and get bounding boxes
crop = page.detection(image)
implt(crop)
bBoxes = words.detection(crop)
# # Simple UI using widgets
# In[6]:
class Cycler:
""" Cycle through the words and recognise them """
height = 60
def __init__(self, image, boxes):
self.boxes = boxes # Array of bounding boxes
self.image = image # Whole image
def recognise(self, img):