def main():
parser = argparse.ArgumentParser()
parser.add_argument('files', metavar="IMAGE_FILE", nargs="+")
parser.add_argument('--threshold', type=float, default=0.90)
args = parser.parse_args()
if len(args.files) % 2 != 0:
parser.error("Files must be provided in pairs")
failures = 0
# TODO: don't punt on media type detection:
for img_1, img_2 in grouper(2, args.files):
digest_1 = image_digest(img_1)
digest_2 = image_digest(img_2)
pcc = cross_correlation(digest_1, digest_2)
if pcc >= args.threshold:
status = 'pass'
log_f = sys.stdout
else:
status = 'FAIL'
log_f = sys.stderr
failures += 1
print('%s\t%s\t%s\t%0.3f' % (status, img_1, img_2, pcc), file=log_f)
return 1 if failures else 0
def compareDirectoryImagesByPhash(self, inputFileName, path2, threshold):
failures=0
total=0
files2 = glob.glob(path2+"/*.jpg")
if (len(files2) == 0):
print("No matching jpg files under "+ path2)
return 0, 0, []
iterations=1
records = len(files2)
log_f = sys.stdout
print('**************** %s%s%s%d%s' % ('Compare ', inputFileName, " images against ", records, path2))
phashList = []
#iterations=10
complete = 1
#complete = iterations / 10
success_pct = 0
#for i, img_1 in enumerate(files):
if (total != 0):
success_pct = (total-failures)*100/total
digest_1 = image_digest(inputFileName)
# if (i % complete == 0):
# print('%s\t%d\t%s\t%d' % ('current iteration ', i, "success_pct", success_pct), file=sys.stdout)
for img_2 in files2:
if (inputFileName == img_2):
continue
#print("Comparing with " + img_2)
total = total + 1
#try:
digest_2 = image_digest(img_2)
pcc = cross_correlation(digest_1, digest_2)
#except:
# continue
if (math.isnan(pcc) == False):
phashList.append((pcc, img_2))
if pcc <= threshold:
status = 'pass'
#log_f = sys.stdout
else:
status = 'fail'
failures += 1
#log_f = sys.stderr
# failures += 1
#print('%s\t%s\t%s\t%0.3f' % (status, inputFileName, img_2, pcc))
else:
print(" Got nan for "+img_2)
failures += 1
success_pct = (total-failures)*100/total
print('%s\t%s\t%d\t%s\t%d\t%s\t%d' % (inputFileName, 'total', total, "failures", failures, ' success_pct ', success_pct))
phashList.sort(reverse = True, key = lambda pcc:pcc[0])
#print(phashList)
return failures, total, phashList
def piece_recognizer(filename):
""" take filename of piece, compare with ref table,
find one of maximum correlation, return ref name """
# piece digest (pd)
pd = image_digest(filename)
# compute cross correlations
ccs = map(lambda ref_im: (ref_im[0], cross_correlation(pd, ref_im[1])), ref_images)
return max(ccs, key=lambda item: item[1])
def detectCard(self):
"""Detect the card from the active frame
"""
# The phash python bindings operate on files, so we have to write our
# current frame to a file to continue
cv2.imwrite('frame.jpg', self.frame)
# Use phash on our frame
ihash = phash.dct_imagehash('frame.jpg')
idigest = phash.image_digest('frame.jpg')
candidates = {}
hashes = self.referencedb.get_hashes()
for MultiverseID in hashes:
if (MultiverseID in self.blacklist):
continue
hamd = phash.hamming_distance(ihash, int(hashes[MultiverseID]))
if (hamd <= self.threshold):
candidates[MultiverseID] = hamd
if (not len(candidates)):
print('No matches found')
return None
finalists = []
minV = min(candidates.values())
for MultiverseID in candidates:
if (candidates[MultiverseID] == minV):
finalists.append(MultiverseID)
bestMatch = None
correlations = {}
for MultiverseID in finalists:
hamd = candidates[MultiverseID]
digest = phash.image_digest(self.referencedb.IMAGE_FILE %
MultiverseID)
corr = phash.cross_correlation(idigest, digest)
if (bestMatch is None or corr > correlations[bestMatch]):
bestMatch = MultiverseID
correlations[MultiverseID] = corr
return bestMatch
def detectCard(self):
"""Detect the card from the active frame
"""
# The phash python bindings operate on files, so we have to write our
# current frame to a file to continue
cv2.imwrite("frame.jpg", self.frame)
# Use phash on our frame
ihash = phash.dct_imagehash("frame.jpg")
idigest = phash.image_digest("frame.jpg")
candidates = {}
hashes = self.referencedb.get_hashes()
for MultiverseID in hashes:
if MultiverseID in self.blacklist:
continue
hamd = phash.hamming_distance(ihash, int(hashes[MultiverseID]))
if hamd <= self.threshold:
candidates[MultiverseID] = hamd
if not len(candidates):
print("No matches found")
return None
finalists = []
minV = min(candidates.values())
for MultiverseID in candidates:
if candidates[MultiverseID] == minV:
finalists.append(MultiverseID)
bestMatch = None
correlations = {}
for MultiverseID in finalists:
hamd = candidates[MultiverseID]
digest = phash.image_digest(self.referencedb.IMAGE_FILE % MultiverseID)
corr = phash.cross_correlation(idigest, digest)
if bestMatch is None or corr > correlations[bestMatch]:
bestMatch = MultiverseID
correlations[MultiverseID] = corr
return bestMatch
def detectCard(self):
"""Detect the card from the active frame
"""
# The phash python bindings operate on files, so we have to write our
# current frame to a file to continue
cv2.imwrite('frame.jpg', self.frame)
# Use phash on our frame
ihash = phash.dct_imagehash('frame.jpg')
idigest = phash.image_digest('frame.jpg')
#print('1a')
candidates = {}
hashes = self.referencedb.get_hashes()
#print('1a1')
#print(hashes)
for MultiverseID in hashes:
#print('id %i' % MultiverseID)
if (MultiverseID in self.blacklist):
continue
hamd = phash.hamming_distance(ihash, int(hashes[MultiverseID]))
#print('1a11')
#print('ham: %i tresh: %i id: %i' % (hamd, self.threshold, MultiverseID))
#print(hamd <= self.threshold)
if (hamd <= self.threshold):
#print('X')
candidates[MultiverseID] = hamd
#print('1a2')
if (not len(candidates)):
print('No matches found')
return None
#print('1a3')
finalists = []
minV = min(candidates.values())
#print('1a4')
for MultiverseID in candidates:
if (candidates[MultiverseID] == minV):
finalists.append(MultiverseID)
#print('1b')
bestMatch = None
correlations = {}
for MultiverseID in finalists:
hamd = candidates[MultiverseID]
#print(self.ROOT_PATH % self.referencedb.IMAGE_FILE % MultiverseID)
s = self.referencedb.IMAGE_FILE % MultiverseID
s = '%s/%s' % (self.ROOT_PATH, s)
#print(s)
#digest = phash.image_digest(self.referencedb.IMAGE_FILE % MultiverseID)
digest = phash.image_digest(s)
corr = phash.cross_correlation(idigest, digest)
if (bestMatch is None or corr > correlations[bestMatch]):
bestMatch = MultiverseID
correlations[MultiverseID] = corr
name, code, rarity = self.referencedb.get_card_info(MultiverseID)
print('Candidate: ' + name + ' [' + code + '] ' + str(corr))
#print('1d')
bestMatches = []
ACCURACY = 1000
print('Finalists:')
print(finalists)
for MultiverseID in finalists:
if correlations[MultiverseID] + ACCURACY > correlations[bestMatch]:
bestMatches.append(MultiverseID)
return bestMatches
#return more finallist
return bestMatch
from phash import image_digest, cross_correlation
from os import path, listdir
def generate_ref_images():
""" canonical piece images used for reference (hence the name) """
piece_colors = ['W','B']
piece_names = ['K','Q','N','B','R','P']
bg_colors = ['L','D']
ref_names = ['_L', '_D'] + map(lambda t: ''.join(t), [(c,p,b) for c in piece_colors for p in piece_names for b in bg_colors])
ref_filenames = map(lambda rn: "ref/%s.jpg" % rn, ref_names)
return ref_filenames
ref_images = map(lambda filename: (filename, image_digest(filename)), generate_ref_images())
def piece_recognizer(filename):
""" take filename of piece, compare with ref table,
find one of maximum correlation, return ref name """
# piece digest (pd)
pd = image_digest(filename)
# compute cross correlations
ccs = map(lambda ref_im: (ref_im[0], cross_correlation(pd, ref_im[1])), ref_images)
return max(ccs, key=lambda item: item[1])
test_inputs = map(lambda s: "tmp/001/%s" % s, listdir("tmp/001"))
for test_input in test_inputs:
name, value = piece_recognizer(test_input)
print "%s = %s (%f match)" % (test_input, name, value)
def compute_hashtable(paths):
return {x:phash.image_digest(x) for x in paths}
