def _swt(pil_img):
pil_img = pil_img.resize((pil_img.size[0] * 4, pil_img.size[1] * 4),
PIL.Image.ANTIALIAS)
pil_img = pillowfight.swt(pil_img, pillowfight.SWT_OUTPUT_BW_TEXT)
pil_img = pil_img.resize(
(int(pil_img.size[0] / 4), int(pil_img.size[1] / 4)),
PIL.Image.ANTIALIAS)
return pil_img
def test_swt(self):
with tempfile.NamedTemporaryFile(suffix='.jpg') as tmpfile:
in_img = PIL.Image.open("tests/data/crappy_background.jpg")
out_img = pillowfight.swt(
in_img, output_type=pillowfight.SWT_OUTPUT_ORIGINAL_BOXES)
in_img.close()
# beware of JPG compression
out_img.save(tmpfile.name)
out_img.close()
out_img = PIL.Image.open(tmpfile.name)
expected_img = PIL.Image.open("tests/data/crappy_background_swt.jpg")
self.assertEqual(out_img.tobytes(), expected_img.tobytes())
expected_img.close()
def test_swt2(self):
with tempfile.NamedTemporaryFile(suffix='.jpg') as tmpfile:
in_img = PIL.Image.open("tests/data/black_border_problem.jpg")
out_img = pillowfight.swt(
in_img, output_type=pillowfight.SWT_OUTPUT_ORIGINAL_BOXES)
in_img.close()
# beware of JPG compression
self.assertEqual(out_img.mode, "RGB")
out_img.save(tmpfile.name)
out_img.close()
out_img = PIL.Image.open(tmpfile.name)
expected_img = PIL.Image.open(
"tests/data/black_border_problem_swt.jpg")
self.assertEqual(out_img.tobytes(), expected_img.tobytes())
expected_img.close()
CNN_model = ConvolutionNN()
# load ground truth
with open(args.gtText, 'r') as gtfile:
gt_data = gtfile.read().replace('\n', '')
# preprocess, GaussianBlur and segmentation
img = cv2.imread(args.image)
# blurred_img = cv2.GaussianBlur(img, (5, 5), 0)
# cv2.imwrite('blurred'+args.image, blurred_img)
# Need to run SWT algorithm to get rid of non text
# with blurred image, it will neglect trivia part, but it will also degrade performance
# PIL_img = Image.open('blurred'+args.image)
PIL_img = Image.open(args.image)
PIL_no_text_img = pf.swt(PIL_img, output_type=pf.SWT_OUTPUT_ORIGINAL_BOXES)
to_extract_img = PIL_to_cv_img(PIL_no_text_img)
# to_extract_img = cv2.imread(args.image, 1)
if DEBUG:
cv2.imshow('img to extract texts', to_extract_img)
cv2.waitKey(0)
cv2.destroyAllWindows()
extracted_string = ''
region_imgs, region_coords = extract_regions(to_extract_img)
for i in range(len(region_imgs)):
lines = extract_words(region_imgs[i])
region_text_block = ''
for words in lines:
im_with_keypoints = cv2.drawKeypoints(img, keypoints, np.array([]), (0,0,255), cv2.DRAW_MATCHES_FLAGS_DRAW_RICH_KEYPOINTS)
# Show keypoints
cv2.imshow("Keypoints", im_with_keypoints)
cv2.waitKey(1)
LIVE_STREAM = "http://192.168.77.81:8080/video"
camera_stream = WebcamVideoStream(src=LIVE_STREAM).start()
framecount = 0
while camera_stream.stream.isOpened():
framecount += 1
frame = cv2.cvtColor(camera_stream.read(), cv2.COLOR_BGR2GRAY)
img = Image.fromarray(frame)
img_out = pillowfight.swt(img, output_type=pillowfight.SWT_OUTPUT_ORIGINAL_BOXES)
print(tesserocr.image_to_text(img))
cv2.imshow("", frame)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
cap.release()
cv2.destroyAllWindows()
#image = cv2.imread("img1.jpg")
#image = cv2.resize(image,(320,240))
#captch_ex(image)
def readImageText(self, imagePath, language="all"):
image = Image.open(imagePath)
image = image.convert("RGB")
image = pillowfight.swt(
image, output_type=pillowfight.SWT_OUTPUT_ORIGINAL_BOXES)
return self.extractText(image, language)
import sys
import PIL
import pillowfight
import pytesseract
img_in = PIL.Image.open(sys.argv[1])
if_natural = sys.argv[2]
if int(if_natural):
img_out = pillowfight.swt(
img_in, output_type=pillowfight.SWT_OUTPUT_GRAYSCALE_TEXT)
img_out.save("results/beforeT", "PNG")
else:
img_out = img_in
config = '-c tessedit_write_images=1'
img_str = pytesseract.image_to_string(img_out, lang='eng', config=config)
print(">>>>>>>>>>>>>>>>>>>>>>>>")
print(img_str)
print("<<<<<<<<<<<<<<<<<<<<<<<<")
from pillowfight import swt
from PIL import Image
import cv2
import numpy
input_image = Image.open("test.jpg")
output_img = swt(input_image)
im = numpy.asarray(output_img)
cv2.imwrite('ouut.jpg', im)
print(output_img)
def main():
parser = argparse.ArgumentParser(description='Replace texts in picture.')
parser.add_argument(
'--ori', help='Original Picture Folder Path', required=True)
parser.add_argument(
'--ref', help='Reference Picture Folder Path', required=True)
parser.add_argument(
'--out', '-o', help='Output folder Path', default='result', metavar='result')
parser.add_argument('--max-feature', type=int,
help='Max feature for aliging', default=5000, metavar='5000')
parser.add_argument('--good-match-percent', type=float,
help='Good match percent for aliging', default=0.01, metavar='0.01')
parser.add_argument('--detect-margin', type=int,
help='margin for detecting replace texts, larger is slower', default=40, metavar='40')
parser.add_argument('--replace-margin', type=int,
help='margin for actually replaced texts', default=20, metavar='20')
args = parser.parse_args()
orifiles = [os.path.join(args.ori, fi) for fi in os.listdir(args.ori)
if fi.endswith('.jpg') or fi.endswith('.png')]
reffiles = [os.path.join(args.ref, fi) for fi in os.listdir(args.ref)
if fi.endswith('.jpg') or fi.endswith('.png')]
i = 0
for ref_filename, im_filename in zip(orifiles, reffiles):
# Read reference image
print("Reading reference image : ", ref_filename)
im_reference = cv2.imread(ref_filename, cv2.IMREAD_COLOR)
# Read image to be aligned
print("Reading image to align : ", im_filename)
im = cv2.imread(im_filename, cv2.IMREAD_COLOR)
print("Aligning images ...")
# Registered image will be resotred in im_aligned.
# The estimated homography will be stored in h.
im_aligned, _ = align_images(
im, im_reference, args.max_feature, args.good_match_percent)
# Write aligned image to disk.
aligned_filename = "aligned.jpg"
print("Saving aligned image : ", aligned_filename)
cv2.imwrite(aligned_filename, im_aligned)
swt_filename = "mask.jpg"
print("Applying swt filter ...")
imp_aligned = Image.open(aligned_filename)
impswt = pillowfight.swt(
imp_aligned, output_type=pillowfight.SWT_OUTPUT_BW_TEXT)
imswt = cv2.cvtColor(np.array(impswt), cv2.COLOR_RGB2BGR)
print("Saving swt filtered image : ", swt_filename)
cv2.imwrite(swt_filename, imswt)
swt_filename_ref = "mask_ref.jpg"
print("Applying swt filter to ref ...")
imp_ref = Image.open(ref_filename)
impswt_ref = pillowfight.swt(
imp_ref, output_type=pillowfight.SWT_OUTPUT_BW_TEXT)
imswt_ref = cv2.cvtColor(np.array(impswt_ref), cv2.COLOR_RGB2BGR)
print("Saving swt filtered image : ", swt_filename_ref)
cv2.imwrite(swt_filename_ref, imswt_ref)
result_filename = os.path.join(args.out, str(i)+".jpg")
print("Replacing texts ...")
imswt_gray = cv2.cvtColor(imswt, cv2.COLOR_BGR2GRAY)
imswt_ref_gray = cv2.cvtColor(imswt_ref, cv2.COLOR_BGR2GRAY)
im_result = replace_images(
im_reference, im_aligned, imswt_gray, imswt_ref_gray,
args.detect_margin, args.replace_margin)
print("Saving result image : ", result_filename)
cv2.imwrite(result_filename, im_result)
im_bound = bound(imswt_gray, imswt_ref_gray)
cv2.imwrite("bound.jpg", im_bound)
i = i+1
"""
Created on Fri Jan 18 20:09:20 2019
@author: Savitoj
"""
import pillowfight
from PIL import Image
from PIL import ImageOps
import cv2
import numpy
import matplotlib.pyplot as plt
input_image = Image.open("New-Folder/img (48).jpg")
#input_image = ImageOps.invert(input_image)
output_img = pillowfight.swt(input_image)
im = numpy.asarray(output_img)
cv2.imwrite("out_img.jpg", im)
mser = cv2.MSER_create()
gray = cv2.cvtColor(im, cv2.COLOR_BGR2GRAY)
for row in range(len(gray)):
for col in range(len(gray[0])):
if row == 0 or col == 0 or row == len(gray) - 1 or col == len(
gray[0]) - 1:
continue
btm = gray[row + 1][col]
top = gray[row - 1][col]
lft = gray[row][col + 1]
right = gray[row][col - 1]
def plate_detected(area, mode, h, w, pre):
output_list = list()
# очистка по предположительному размеру плейта
# сейчас идет расчет относительно размера найденой фичи (лицо, рост, профиль)
# возможно лучше сделать относительно предпологаемой области нахождения плейта
'''if mode == 'profiles' or mode == 'faces':
# параметры для расчета относительно размера фичи
min_expected_width = h / 4
min_expected_height = h / 4
max_expected_width = h * 1.5
max_expected_height = h * 1.5
elif mode == 'fullbodys':
min_expected_width = h
min_expected_height = h
max_expected_width = h
max_expected_height = h'''
# относительно рамера предпологаемой области плейта
# одинаковые значения для всех фичь, так как размер предпологаемой области почти одинаков
min_expected_width = int(w / 6)
min_expected_height = int(h / 15)
max_expected_width = int(w)
max_expected_height = int(h / 3)
# cv2.imshow('area', area)
# cv2.waitKey(0)
new = Image.fromarray(area)
img_out = pillowfight.swt(
new, output_type=pillowfight.SWT_OUTPUT_ORIGINAL_BOXES)
# img_out.show()
img_out = np.array(img_out)
img_out = cv2.cvtColor(img_out, cv2.COLOR_BGR2GRAY)
# cv2.imshow('img_out', img_out)
# cv2.waitKey(0)
# черные блоки областей возвращаемх pillowfight.swt
area_plate_block = cv2.threshold(img_out, 254, 255,
cv2.THRESH_BINARY_INV)[1]
# cv2.imshow('black_box', area_plate_block)
# cv2.waitKey(0)
contours, hierarchy = cv2.findContours(area_plate_block, cv2.RETR_TREE,
cv2.CHAIN_APPROX_SIMPLE)
# отрисовка контуров черных блоков из pillowfight.swt
# arr = np.zeros([area.shape[0], area.shape[1]])
# arr = cv2.drawContours(arr, contours, -1, 255)
# cv2.imshow('black_box_contour', arr)
# cv2.waitKey(0)
# структура contour[0][0][0] : первый элемент - индес точки(против чавовой начиная с левой верхней)
# второй - всегда 0, третий - 0->x 1->y
for contour in contours:
if len(contour) == 4:
width = contour[3][0][0] - contour[0][0][0]
height = contour[1][0][1] - contour[0][0][1]
if (min_expected_width < width < max_expected_width)\
and (min_expected_height < height < max_expected_height):
y1 = contour[0][0][1]
y2 = contour[1][0][1]
x1 = contour[0][0][0]
x2 = contour[3][0][0]
width_extension = 12
height_extension = 8
if y1 - height_extension < 0:
y1 = 0
else:
y1 = y1 - height_extension
if y2 + height_extension > area.shape[0]:
y2 = area.shape[0]
else:
y2 = y2 + height_extension
if x1 - width_extension < 0:
x1 = 0
else:
x1 = x1 - width_extension
if x2 + width_extension > area.shape[1]:
x2 = area.shape[1]
else:
x2 = x2 + width_extension
expected_plate = area[y1:y2, x1:x2]
output_list.append(plate_symbol_detected(expected_plate, pre))
clear_output_list = [x for x in output_list if x is not None]
if len(clear_output_list) > 0:
print(clear_output_list)
return clear_output_list
else:
return None
def SWT(Img, name):
#img_in = PIL.Image.open(name+extensao)
img_out = pillowfight.swt(
Img, output_type=pillowfight.SWT_OUTPUT_GRAYSCALE_TEXT)
img_out.save(path + name + " - 7SWT" + extensao)
def main_func(image_path):
#start
PATH_ORIGINAL = image_path
originalPath = PATH_ORIGINAL
imgSkewed = Image.open(originalPath)
imgSwt = pillowfight.swt(imgSkewed, output_type=pillowfight.SWT_OUTPUT_BW_TEXT)
imgSwt.save(PATH_SWT)
# saved skewed Image
# imageCopied = convertPilImageToCv2Image(imgSwt)
image = convertPilImageToCv2Image(imgSwt)
results, image = east(image)
# TODO
rect = results[-4][0]
# import pdb
# pdb.set_trace()
roi = image[rect[1]-40:rect[3]+40, rect[0]-40:rect[2]+40]
if roi==[]:
roi = image[rect[1]-4:rect[3], rect[0]-4:rect[2]]
# cv2.imwrite(PATH_CROP_AFTER_EAST, roi)
# roi = image[rect[1]-4:rect[3], rect[0]-4:rect[2]]
image = roi
# for rect in results[-2]:
# roi = image[rect[1]-40:rect[3]+40, rect[0]-40:rect[2]+40]
# cv2.imwrite('crop.jpg',roi)
# break
#################################################################################################
# convert the image to grayscale and flip the foreground
# and background to ensure foreground is now "white" and
# the background is "black"
gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
gray = cv2.bitwise_not(gray)
# threshold the image, setting all foreground pixels to
# 255 and all background pixels to 0
thresh = cv2.threshold(gray, 0, 255,
cv2.THRESH_BINARY | cv2.THRESH_OTSU)[1]
# grab the (x, y) coordinates of all pixel values that
# are greater than zero, then use these coordinates to
# compute a rotated bounding box that contains all
# coordinates
coords = np.column_stack(np.where(thresh > 0))
angle = cv2.minAreaRect(coords)[-1]
# the `cv2.minAreaRect` function returns values in the
# range [-90, 0); as the rectangle rotates clockwise the
# returned angle trends to 0 -- in this special case we
# need to add 90 degrees to the angle
if angle < -45:
angle = -(90 + angle)
# otherwise, just take the inverse of the angle to make
# it positive
else:
angle = -angle
# print("angle " + str(angle))
# rotate the image to deskew it
(h, w) = image.shape[:2]
center = (w // 2, h // 2)
M = cv2.getRotationMatrix2D(center, angle, 1.0)
img = cv2.imread(PATH_ORIGINAL)
(h, w) = img.shape[:2]
rotated = cv2.warpAffine(img, M, (w, h),
flags=cv2.INTER_CUBIC, borderMode=cv2.BORDER_REPLICATE)
# # draw the correction angle on the image so we can validate it
# cv2.putText(rotated, "Angle: {:.2f} degrees".format(angle),
# (10, 30), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (0, 0, 255), 2)
# show the output image
# print("[INFO] angle: {:.3f}".format(angle))
# cv2.imshow("Input", image)
# cv2.imshow("Rotated", rotated)
# cv2.imwrite(PATH_ORIGINAL_ROTATED,rotated)
# cv2.waitKey(0)
#################################################################
image = convertCv2ImageToPilImage(rotated)
image.save(PATH_300_DPI_IMAGE,dpi=(300,300))
results, image = east(convertPilImageToCv2Image(image))
# print(results)
######################################################################################################33\
imgSkewed = Image.open(PATH_300_DPI_IMAGE)
# imgSwt = pillowfight.swt(imgSkewed, output_type=pillowfight.SWT_OUTPUT_BW_TEXT)
imgSkewed.save(PATH_SWT)
cmd = "tesseract " + PATH_SWT + " test -l eng --psm 11 --oem 1"
returned_value = os.system(cmd)
# returns the exit code in unix
text = open("test.txt", "r")
text1 = str(text.read())
print(text.read())
return (text1)
import sys
import pillowfight
import pyocr
import pyocr.builders
tools = pyocr.get_available_tools()
if len(tools) == 0:
print("No OCR tool found")
sys.exit(1)
# The tools are returned in the recommended order of usage
tool = tools[0]
print("Will use tool '%s'" % (tool.get_name()))
# Ex: Will use tool 'libtesseract'
langs = tool.get_available_languages()
print("Available languages: %s" % ", ".join(langs))
lang = 'eng'
print("Will use lang '%s'" % (lang))
# Ex: Will use lang 'fra'
# Note that languages are NOT sorted in any way. Please refer
# to the system locale settings for the default language
# to use.
img_in = PIL.Image.open(sys.argv[1])
img_out = pillowfight.swt(img_in, output_type=pillowfight.SWT_OUTPUT_BW_TEXT)
txt = tool.image_to_string(Image.open(sys.argv[1]),
lang=lang,
builder=pyocr.builders.TextBuilder())
print(">>>>>>>>>>>>>>")
print(txt)
print("<<<<<<<<<<<<<<")
out = img.resize(tuple(i * 2 for i in img.size), PIL.Image.LANCZOS)
process = True
if process:
out = pillowfight.ace(img, slope=10, limit=1000, samples=100, seed=None)
# Canny edge detection
# out = pillowfight.canny(out)
# Gaussian blur
# out = pillowfight.gaussian(out, sigma=2.0, nb_stddev=5)
# sobel
# out = pillowfight.sobel(out)
# Stroke Width Transform
# SWT_OUTPUT_BW_TEXT = 0 # default
# SWT_OUTPUT_GRAYSCALE_TEXT = 1
# SWT_OUTPUT_ORIGINAL_BOXES = 2
out = pillowfight.swt(out, output_type=2)
#unpaper
out = pillowfight.unpaper_blackfilter(out)
out = pillowfight.unpaper_blurfilter(out)
out = pillowfight.unpaper_border(out)
# out = pillowfight.unpaper_grayfilter(out) #adds white rectangles?
out = pillowfight.unpaper_masks(out)
out = pillowfight.unpaper_noisefilter(out)
print(pytesseract.image_to_string(out))
out.save("out.png")