import ocr_utils
foia_response = '1505deposists2.pdf'
ocr_utils.split_pdf(foia_response)
num_page_start = 1
num_page_end = 86
all_text = []
data=[]
for page in range(num_page_start, num_page_end + 1):
indiv_filename = "{}.{}.pdf".format(ocr_utils.get_stem(foia_response), page)
tmp_filename = indiv_filename
tmp_image = ocr_utils.convert_to_png(tmp_filename)
all_text.append(ocr_utils.ocr_image(tmp_image))
data=all_text[0].split('\n')
# for each data set you should read in the amounts
reviewed=[]
# inputs moved to checkanalysis.py
#for row in data:
#
# if row.split() == []: pass
# else:
# inputs.append(row.split()[2])
#print inputs
draw2 = np.copy(im_resized)
boxes = get_boxes(segm, rbox, angle_pred, args.segm_thresh)
img = Image.fromarray(draw2)
draw = ImageDraw.Draw(img)
# if len(boxes) > 10:
# boxes = boxes[0:10]
out_boxes = []
for box in boxes:
pts = box[0:8]
pts = pts.reshape(4, -1)
det_text, conf, dec_s = ocr_image(net, codec, im_data, box,
device)
if len(det_text) == 0:
continue
width, height = draw.textsize(det_text, font=font2)
center = [box[0], box[1]]
draw.text((center[0], center[1]),
det_text,
fill=(0, 255, 0),
font=font2)
out_boxes.append(box)
print(det_text)
im = np.array(img)
for box in out_boxes:
pts = box[0:8]
foia_response = '1505deposists2.pdf'
ocr_utils.split_pdf(foia_response)
num_page_start = 1
num_page_end = 86
all_text = []
data = []
for page in range(num_page_start, num_page_end + 1):
indiv_filename = "{}.{}.pdf".format(ocr_utils.get_stem(foia_response),
page)
tmp_filename = indiv_filename
tmp_image = ocr_utils.convert_to_png(tmp_filename)
all_text.append(ocr_utils.ocr_image(tmp_image))
data = all_text[0].split('\n')
# for each data set you should read in the amounts
reviewed = []
# inputs moved to checkanalysis.py
#for row in data:
#
# if row.split() == []: pass
# else:
# inputs.append(row.split()[2])
#print inputs
def run_model_input_image(im, show_boxes=False):
predictions = {}
parser = argparse.ArgumentParser()
parser.add_argument('-cuda', type=int, default=1)
parser.add_argument('-model', default='e2e-mlt-rctw.h5')
parser.add_argument('-segm_thresh', default=0.5)
font2 = ImageFont.truetype("Arial-Unicode-Regular.ttf", 18)
args = parser.parse_args()
net = ModelResNetSep2(attention=True)
net_utils.load_net(args.model, net)
net = net.eval()
if args.cuda:
print('Using cuda ...')
net = net.cuda()
with torch.no_grad():
# im = Image.open(im)
# im = im.convert('RGB')
im = np.asarray(im)
im = im[...,:3]
im_resized, (ratio_h, ratio_w) = resize_image(im, scale_up=False)
images = np.asarray([im_resized], dtype=np.float)
images /= 128
images -= 1
im_data = net_utils.np_to_variable(images, is_cuda=args.cuda).permute(0, 3, 1, 2)
seg_pred, rboxs, angle_pred, features = net(im_data)
rbox = rboxs[0].data.cpu()[0].numpy()
rbox = rbox.swapaxes(0, 1)
rbox = rbox.swapaxes(1, 2)
angle_pred = angle_pred[0].data.cpu()[0].numpy()
segm = seg_pred[0].data.cpu()[0].numpy()
segm = segm.squeeze(0)
draw2 = np.copy(im_resized)
boxes = get_boxes(segm, rbox, angle_pred, args.segm_thresh)
img = Image.fromarray(draw2)
draw = ImageDraw.Draw(img)
#if len(boxes) > 10:
# boxes = boxes[0:10]
out_boxes = []
prediction_i = []
for box in boxes:
pts = box[0:8]
pts = pts.reshape(4, -1)
det_text, conf, dec_s = ocr_image(net, codec, im_data, box)
if len(det_text) == 0:
continue
width, height = draw.textsize(det_text, font=font2)
center = [box[0], box[1]]
draw.text((center[0], center[1]), det_text, fill = (0,255,0),font=font2)
out_boxes.append(box)
# det_text is one prediction
prediction_i.append(det_text.lower())
predictions["frame"] = prediction_i
# show each image boxes and output in pop up window.
show_image_with_boxes(img, out_boxes, show=show_boxes)
print(predictions)
return predictions
# PNGs created by running the image through an image converter
# i.e. convert -density 300 -trim <file>.pdf -quality 100 -sharpen 0x1.0 M <image>.png
# density > 250 leads to poor image recogition.
# depends on tesseract-ocr and tesseract-devvel nd leptonica to run properly
import ocr_utils
import numpy
png_response = '1505deposists2.83.png'
all_text=[]
all_text.append(ocr_utils.ocr_image(png_response))
data=all_text[0].split('\n')
# for each data set you should read in the amounts
# row.split()[2] should work for *most* cases but there are
# edge cases where [0] is required i.e. 1505deposits2.8.png
inputs=[]
for row in data:
if row.split() == []: pass
try:
inputs.append(row.split()[2])
except IndexError:
inputs.append('null')
# check which checks require review and mark them for review
review=[]
# create new array and flag which amounts require review
# afterwards, remove paranthesis to get purely dollar amounts
