def processImg(img, template_zero, template_one, imgpath):
""" The pipeline for processing one image:
1) crop out two rough barcode regions from the image
2) run template matching against it with two templates with criteria,
retrieving the best matches
3) process matching result, transform into 01-bitstring
Note: Only the front-side has a full barcodes on the UL/UR corners.
The back-side, however, has "0", "" on the top, and "0", "0" on the
bottom. We can leverage this information.
Output:
list DECODINGS, dict MARKS_OUT.
list DECODINGS: [str decoding_upperLeft, str decoding_upperRight]
dict MARKS_OUT: maps {MARK_ON/MARK_OFF: [(imgpath, (x1,y1,x2,y2), LEFT/RIGHT), ...]}
"""
w_img, h_img = cv.GetSize(img)
w_fact, h_fact = 0.15, 0.20
w_patch, h_patch = int(round(w_img * w_fact)), int(round(h_img * h_fact))
x_off, y_off = 0, 10
cv.SetImageROI(img, (x_off, y_off, w_patch, h_patch))
patchLeft = tempmatch.smooth(img, 3, 3, bordertype='const', val=255.0)
decodings_left, zero_locs_left, one_locs_left = decode_patch(
patchLeft, template_zero, template_one, imgpath)
cv.ResetImageROI(img)
marks_out_left = create_marks_dict(zero_locs_left, one_locs_left, x_off,
y_off, imgpath, LEFT)
x_off = int(round(w_img - w_patch))
cv.SetImageROI(img, (x_off, y_off, w_patch, h_patch))
patchRight = tempmatch.smooth(img, 3, 3, bordertype='const', val=255.0)
decodings_rht, zero_locs_rht, one_locs_rht = decode_patch(
patchRight, template_zero, template_one, imgpath)
cv.ResetImageROI(img)
marks_out_rht = create_marks_dict(zero_locs_rht, one_locs_rht, x_off,
y_off, imgpath, RIGHT)
marks_out = marks_out_left
for markval, tups in marks_out_rht.iteritems():
marks_out.setdefault(markval, []).extend(tups)
return (decodings_left, decodings_rht), marks_out
sequoia.ORIG_IMG_H, exmpl_imgsize[0],
exmpl_imgsize[1])
IsymB = sequoia.rescale_img(IsymB, sequoia.ORIG_IMG_W,
sequoia.ORIG_IMG_H, exmpl_imgsize[0],
exmpl_imgsize[1])
IsymC = sequoia.rescale_img(IsymC, sequoia.ORIG_IMG_W,
sequoia.ORIG_IMG_H, exmpl_imgsize[0],
exmpl_imgsize[1])
IsymD = sequoia.rescale_img(IsymD, sequoia.ORIG_IMG_W,
sequoia.ORIG_IMG_H, exmpl_imgsize[0],
exmpl_imgsize[1])
IsymE = sequoia.rescale_img(IsymE, sequoia.ORIG_IMG_W,
sequoia.ORIG_IMG_H, exmpl_imgsize[0],
exmpl_imgsize[1])
Itemp0 = tempmatch.smooth(Itemp0, 3, 3, bordertype='const', val=255.0)
Itemp1 = tempmatch.smooth(Itemp1, 3, 3, bordertype='const', val=255.0)
IsymA = tempmatch.smooth(IsymA, 3, 3, bordertype='const', val=255.0)
IsymB = tempmatch.smooth(IsymB, 3, 3, bordertype='const', val=255.0)
IsymC = tempmatch.smooth(IsymC, 3, 3, bordertype='const', val=255.0)
IsymD = tempmatch.smooth(IsymD, 3, 3, bordertype='const', val=255.0)
IsymE = tempmatch.smooth(IsymE, 3, 3, bordertype='const', val=255.0)
backsmap = {} # maps {ballotid: [backpath_i, ...]}
def mygauge_tick(N=1):
""" Add N ticks to the progress bar. """
for _ in xrange(N):
queue.put(True)
if num_pages == 1:
return handle_singleside(ballots, Itemp0, Itemp1, IsymA, IsymB, IsymC,
def main():
args = sys.argv[1:]
arg0 = args[0]
do_display_output = True if 'display' in args else False
if do_display_output:
outdir = args[-1]
else:
outdir = None
do_profile = True if 'profile' in args else False
if os.path.isdir(arg0):
imgpaths = []
for dirpath, dirnames, filenames in os.walk(arg0):
for imgname in [f for f in filenames if isimgext(f)]:
imgpaths.append(os.path.join(dirpath, imgname))
else:
imgpaths = [arg0]
template_zero_path = "sequoia_template_zero_skinny.png"
template_one_path = "sequoia_template_one_skinny.png"
sidesymbol_path = "sequoia_side_symbol.png"
Izero = cv.LoadImage(template_zero_path, cv.CV_LOAD_IMAGE_GRAYSCALE)
Ione = cv.LoadImage(template_one_path, cv.CV_LOAD_IMAGE_GRAYSCALE)
Isidesym = cv.LoadImage(sidesymbol_path, cv.CV_LOAD_IMAGE_GRAYSCALE)
IsymA = cv.LoadImage(SYMA_IMGPATH, cv.CV_LOAD_IMAGE_GRAYSCALE)
IsymB = cv.LoadImage(SYMB_IMGPATH, cv.CV_LOAD_IMAGE_GRAYSCALE)
IsymC = cv.LoadImage(SYMC_IMGPATH, cv.CV_LOAD_IMAGE_GRAYSCALE)
IsymD = cv.LoadImage(SYMD_IMGPATH, cv.CV_LOAD_IMAGE_GRAYSCALE)
IsymE = cv.LoadImage(SYME_IMGPATH, cv.CV_LOAD_IMAGE_GRAYSCALE)
# Rescale IZERO/IONE/ISIDESYM to match this dataset's image dimensions
exmpl_imgsize = cv.GetSize(cv.LoadImage(imgpaths[0]))
if exmpl_imgsize != (ORIG_IMG_W, ORIG_IMG_H):
print "...rescaling images..."
Izero = rescale_img(Izero, ORIG_IMG_W, ORIG_IMG_H, exmpl_imgsize[0],
exmpl_imgsize[1])
Ione = rescale_img(Ione, ORIG_IMG_W, ORIG_IMG_H, exmpl_imgsize[0],
exmpl_imgsize[1])
Isidesym = rescale_img(Isidesym, ORIG_IMG_W, ORIG_IMG_H,
exmpl_imgsize[0], exmpl_imgsize[1])
IsymA = rescale_img(IsymA, ORIG_IMG_W, ORIG_IMG_H, exmpl_imgsize[0],
exmpl_imgsize[1])
IsymB = rescale_img(IsymB, ORIG_IMG_W, ORIG_IMG_H, exmpl_imgsize[0],
exmpl_imgsize[1])
IsymC = rescale_img(IsymC, ORIG_IMG_W, ORIG_IMG_H, exmpl_imgsize[0],
exmpl_imgsize[1])
IsymD = rescale_img(IsymD, ORIG_IMG_W, ORIG_IMG_H, exmpl_imgsize[0],
exmpl_imgsize[1])
IsymE = rescale_img(IsymE, ORIG_IMG_W, ORIG_IMG_H, exmpl_imgsize[0],
exmpl_imgsize[1])
Izero = tempmatch.smooth(Izero, 3, 3, bordertype='const', val=255.0)
Ione = tempmatch.smooth(Ione, 3, 3, bordertype='const', val=255.0)
Isidesym = tempmatch.smooth(Isidesym, 3, 3, bordertype='const', val=255.0)
IsymA = tempmatch.smooth(IsymA, 3, 3, bordertype='const', val=255.0)
IsymB = tempmatch.smooth(IsymB, 3, 3, bordertype='const', val=255.0)
IsymC = tempmatch.smooth(IsymC, 3, 3, bordertype='const', val=255.0)
IsymD = tempmatch.smooth(IsymD, 3, 3, bordertype='const', val=255.0)
IsymE = tempmatch.smooth(IsymE, 3, 3, bordertype='const', val=255.0)
t = time.time()
err_imgpaths = []
for imgpath in imgpaths:
I = cv.LoadImage(imgpath, cv.CV_LOAD_IMAGE_GRAYSCALE)
print "For imgpath {0}:".format(imgpath)
side, isflip = get_side(I, IsymA, IsymB, IsymC, IsymD, IsymE)
if side == None:
print " ERROR GET_SIDE"
err_imgpaths.append(imgpath)
continue
elif side == 1:
print " Detected Backside, isflip={0}".format(isflip)
continue
cv.ResetImageROI(I)
if isflip:
cv.Flip(I, I, flipMode=-1)
decodings, marklocs = decode(I, Izero, Ione, _imgpath=imgpath)
if decodings == None:
print " ERROR DECODE"
err_imgpaths.append(imgpath)
continue
else:
print " {0} isflip={1}".format(decodings, isflip)
if not do_display_output:
continue
# Output colorful image with interpretation displayed nicely
Icolor = cv.LoadImage(imgpath, cv.CV_LOAD_IMAGE_COLOR)
if isflip:
cv.Flip(Icolor, Icolor, flipMode=-1)
for marktype, tups in marklocs.iteritems():
if marktype == MARK_ON:
color = cv.CV_RGB(0, 0, 255)
else:
color = cv.CV_RGB(255, 0, 0)
for (imgpath, (x1, y1, x2, y2), userdata) in tups:
cv.Rectangle(Icolor, (x1, y1), (x2, y2), color, thickness=2)
imgname = os.path.split(imgpath)[1]
outrootdir = os.path.join(outdir, imgname)
try:
os.makedirs(outrootdir)
except:
pass
outpath = os.path.join(
outrootdir, "{0}_bbs.png".format(os.path.splitext(imgname)[0]))
cv.SaveImage(outpath, Icolor)
dur = time.time() - t
print "...Finished Decoding {0} images ({1} s).".format(len(imgpaths), dur)
print " Avg. Time per Image: {0} s".format(dur / float(len(imgpaths)))
print " Number of Errors: {0}".format(len(err_imgpaths))
print "Done."