def createVPTree(images, hashfile, treefile):
imagePaths = list(paths.list_images(images))
hashes = {}
for (i, imagePath) in enumerate(imagePaths):
# print("[INFO] processing image {}/{}".format(i + 1,
# len(imagePaths)))
image = cv2.imread(imagePath)
# compute the hash for the image and convert it
h = dhash(image)
h = convert_hash(h)
# update the hashes dictionary
l = hashes.get(h, [])
l.append(imagePath)
hashes[h] = l
# build the VP-Tree
print("[INFO] building VP-Tree...")
points = list(hashes.keys())
tree = vptree.VPTree(points, hammingDistance)
# serialize the VP-Tree to disk
print("[INFO] serializing VP-Tree...")
f = open(treefile, "wb")
f.write(pickle.dumps(tree))
f.close()
# serialize the hashes to dictionary
print("[INFO] serializing hashes...")
f = open(hashfile, "wb")
f.write(pickle.dumps(hashes))
f.close()
def trackPerson(image, hashes, tree, distance=15):
global PERSON_COUNT
# compute the hash for the image and convert it
h = dhash(image)
h = convert_hash(h)
person_id = -1
# search for similar person
if not tree == None:
sim = sorted(tree.get_all_in_range(h, distance))
if len(sim):
p_ids = hashes.get(sim[0][1], [])
if len(p_ids):
person_id = p_ids[0]
if person_id == -1:
PERSON_COUNT += 1
person_id = PERSON_COUNT
# update the hashes dictionary
l = hashes.get(h, [])
l.append(person_id)
hashes[h] = l
points = list(hashes.keys())
tree = vptree.VPTree(points, hammingDistance)
return tree, hashes, person_id
def index_person(self, image, distance=15):
# compute the hash for the image and convert it
h = dhash(image)
h = convert_hash(h)
person_id = -1
# search for similar person
if self.tree is not None:
sim = sorted(tree.get_all_in_range(h, distance))
if len(sim):
p_ids = self.hashes.get(sim[0][1], [])
if len(p_ids):
person_id = p_ids[0]
if person_id == -1:
PERSON_COUNT += 1
person_id = PERSON_COUNT
# update the hashes dictionary
l = self.hashes.get(h, [])
l.append(person_id)
self.hashes[h] = l
points = list(hashes.keys())
self.tree = vptree.VPTree(points, hammingDistance)
return person_id
def compute_hashes(img_paths, hashes={}):
for img_path in img_paths:
hashed = hash_distance(img_path)
idx = convert_hash(hashed)
loc = hashes.get(idx, [])
loc.append(img_path)
hashes[idx] = loc
return hashes
default=100,
help="maximum hamming distance")
args = vars(ap.parse_args())
# load the VP-Tree and hashes dictionary
print("[INFO] loading VP-Tree and hashes...")
tree = pickle.loads(open(args["tree"], "rb").read())
hashes = pickle.loads(open(args["hashes"], "rb").read())
# load the input query image
image = cv2.imread(args["query"])
cv2.imshow("Query", image)
# compute the hash for the query image, then convert it
queryHash = dhash(image)
queryHash = convert_hash(queryHash)
def searchSimilarImages(distance):
# perform the search
print("[INFO] performing search...")
start = time.time()
results = tree.get_all_in_range(queryHash, distance)
results = sorted(results)
end = time.time()
print("[INFO] search took {} seconds".format(end - start))
return results
def main():
help="maximum hamming distance")
# ^^^^ ADJUST THIS METRIC FOR QUERY THRESHOLD (larger distance=more images to compare [longer runtime])
ap.add_argument("-s",
"--size",
required=False,
type=str,
help="image resize (default is 8x8)")
args = vars(ap.parse_args())
# load the input query image
image = cv2.imread(args["query"])
#cv2.imshow("Query", image)
# compute the hash for the query image, then convert it
queryHash = hs.dhash(image, int(args["size"]))
queryHash = hs.convert_hash(queryHash)
print("the query image hash value is", queryHash)
# load the VP-Tree and hashes dictionary
print("[INFO] loading VP-Tree and hashes...")
#tree=pickle.loads(open(args["tree"], "rb").read())
hashes = pickle.loads(open(args["hashes"], "rb").read())
start = time.time()
resultsList = [] #Adds results of image query to this list
for pickleTree in glob.glob(args["tree"] + "/vptree_*.pickle"):
print("[INFO] loading VP-Tree: {pickle}".format(pickle=pickleTree))
with open(pickleTree, 'rb') as f:
tree = pickle.load(f)
#tree=pickle.loads(open(pickleTree, "rb").read())
folders='./'
#Import as Spark RDD
urlsRDD=sc.textFile("s3a://"+bucket+"/urls.txt")
#llist = urlsRDD.collect()
#urlsRDD.take(100).foreach(println)
#print(urlsRDD)
#impg.read_image_from_s3(bucket, url)
#Download and acquire image vectors
img_vectors=urlsRDD.map(lambda url: (url, impg.read_image_from_s3(bucket, url)))
#img_vectors.take(5)
#dHash function
img_hash=img_vectors.map(lambda img: (img[0], hs.convert_hash(hs.dhash(img[1], 32))))
#Makes dictionary from RDD continaing dHash (key) and URLs (value)
#dHash_dict=img_hash.map(lambda (url, dHash): (dHash, url)) ### python 2 code
dHash_dict=img_hash.map(lambda url_dHash: (url_dHash[1], url_dHash[0])) ### python 3 code
#dHash_dict.take(5).foreach(println)
#Pickles python hash dictionary
hs.pickleHash(dHash_dict.collectAsMap())
#Converts Image dHash into Sparse Vector (Required Input for LSH)
img_sparse=img_hash.map(lambda img: (img[0], str(img[1]), hs.sparse_vectorize(img[1])))
#Converts array of sparse img vectors into dataframe
import pickle
#S3 Bucket/Folder
bucket = 'vasco-imagenet-db'
folders = 'test_small'
#Import as Spark RDD
urlsRDD = sc.textFile("s3a://" + bucket + "/urls.txt")
#Download and acquire image vectors
img_vectors = urlsRDD.map(lambda url:
(url, impg.read_image_from_s3(bucket, url)))
#dHash function
img_hash = img_vectors.map(lambda img:
(img[0], hs.convert_hash(hs.dhash(img[1], 32))))
#Makes dictionary from RDD continaing dHash (key) and URLs (value)
dHash_dict = img_hash.map(lambda (url, dHash): (dHash, url))
#Pickles python hash dictionary
hs.pickleHash(dHash_dict.collectAsMap())
#Converts Image dHash into Sparse Vector (Required Input for LSH)
img_sparse = img_hash.map(lambda img:
(img[0], str(img[1]), hs.sparse_vectorize(img[1])))
#Converts array of sparse img vectors into dataframe
df = spark.createDataFrame(img_sparse, ["url", "dHash", "sparseHash"])
#MinHashLSH
print(f" {i}")
cv2.waitKey(0)
def check(img_path):
if not os.path.isfile(img_path):
sys.exit(1)
filename, ext = os.path.splitext(img_path)
if ext not in [".jpg", ".jpeg", ".png", ".bmp"]:
sys.exit(1)
return img_path
if __name__ == "__main__":
ap = argparse.ArgumentParser()
ap.add_argument("-u", "--upload", required=True, type=str)
args = vars(ap.parse_args())
img_path = check(args["upload"])
tree = pickle.loads(open("vptree.pickle", "rb").read())
hashes = pickle.loads(open("hashes.pickle", "rb").read())
query_hash = hash_distance(img_path)
query_idx = convert_hash(query_hash)
search_results = search(tree, query_idx)
img_paths = get_imgs(search_results, hashes)
display(img_paths)
##
def search():
if request.method == "POST":
RESULTS_ARRAY = []
# get url
image_url = request.form.get('img')
print(image_url)
try:
# download image into array (from url)
if 'http' in image_url:
resp = urllib.request.urlopen(image_url)
image = np.asarray(bytearray(resp.read()), dtype="uint8")
image = cv2.imdecode(image, cv2.IMREAD_COLOR)
else:
# load the input query image (from webserver folder)
image = cv2.imread('.' + image_url.split('..')[-1])
# compute the hash for the query image, then convert it
queryHash = hs.dhash(image,
32) #manually change to match indexed iumages
queryHash = hs.convert_hash(queryHash)
# load the VP-Tree and hashes dictionary
print("[INFO] loading VP-Tree and hashes...")
hashes = pickle.loads(
open('static/pickles/img_hash_dictionary.pickle', "rb").read())
start = time.time()
resultsList = [] #Adds results of image query to this list
for pickleTree in glob.glob("static/pickles/vptree_*.pickle"):
#print("[INFO] loading VP-Tree: {pickle}".format(pickle=pickleTree))
with open(pickleTree, 'rb') as f:
tree = pickle.load(f)
#Perform search in VPTree
#print("[INFO] performing search on {pickle}".format(pickle=pickleTree))
results = tree.get_all_in_range(
queryHash, 50
) #Tune to lower computational time but yield at least four results
results = sorted(results)
#Loop through reults and add to resultsList
counter = 0 #Ensure that only top 10 results are used
for i, result in enumerate(results):
resultsList.append(result)
if i >= 1:
break #Grabs first result (modifiable), moves on to next tree
else:
i += 1
#Sort final list of all resutls
resultsList = sorted(resultsList)
end = time.time()
print("[INFO] search took {} seconds".format(end - start))
# loop over the results
for (score, h) in resultsList[:10]:
#grab all image paths in our dataset with the same hash
resultPaths = [hashes.get(int(h), [])]
print("[INFO] {} total images(s) with d: {}, h:{}".format(
len(resultPaths), score, h))
# loop over the result paths
for resultID in resultPaths:
#Remove URL Path Prefix (prefix is already included in output)
#resultID=str(resultID).split('/')[-1]
# print(resultID)
# load the result image and display it to our screeni
# RESULTS_ARRAY.append(
# {"image": str(resultID), "score": str(score)})
# print(RESULTS_ARRAY)
RESULTS_ARRAY.append({
"image":
'http://vasco-imagenet-db.s3-us-west-2.amazonaws.com/'
+ str(resultID),
"score":
str(score)
})
#Change the bucket to match what is being queired and change view permissions
# return success
print(RESULTS_ARRAY)
return jsonify(results=(RESULTS_ARRAY[:4]))
except:
# return error
#return jsonify({"sorry": "Sorry, no results! Please try again."}), 500
raise
