Find duplicate images in fastest way - python

I have 2 image folder containing 10k and 35k images. Each image is approximately the size of (2k,2k).
I want to remove the images which are exact duplicates.
The variation in different images are just a change in some pixels.
I have tried DHashing, PHashing, AHashing but as they are lossy image hashing technique so they are giving the same hash for non-duplicate images too.
I also tried writing a code in python, which will just subtract images and the combination in which the resultant array is not zero everywhere gives those image pair to be duplicate of each other.
Buth the time for a single combination is 0.29 seconds and for total 350 million combinations is really huge.
Is there a way to do it in a faster way without flagging non-duplicate images also.
I am open to doing it in any language(C,C++), any approach(distributed computing, multithreading) which can solve my problem accurately.
Apologies if I added some of the irrelevant approaches as I am not from computer science background.
Below is the code I used for python approach -
start = timeit.default_timer()
dict = {}
for i in path1:
img1 = io.imread(i)
base1 = os.path.basename(i)
for j in path2:
img2 = io.imread(j)
base2 = os.path.basename(j)
if np.array_equal(img1, img2):
err = img1.astype('float') - img2.astype('float')
is_all_zero = np.all((err == 0))
if is_all_zero:
dict[base1] = base2
else:
continue
stop = timeit.default_timer()
print('Time: ', stop - start)

Use lossy hashing as a prefiltering step, before a complete comparison. You can also generate thumbnail images (say 12 x 8 pixels), and compare for similarity.
The idea is to perform quick rejection of very different images.

You should find the answer on how to delete duplicate files (not only images). Then you can use, for example, fdupes or find some alternative SW: https://alternativeto.net/software/fdupes/

This code checks if there are any duplicates in a folder (it's a bit slow though):
import image_similarity_measures
from image_similarity_measures.quality_metrics import rmse, psnr
from sewar.full_ref import rmse, psnr
import cv2
import os
import time
def check(path_orginal,path_new):#give r strings
original = cv2.imread(path_orginal)
new = cv2.imread(path_new)
return rmse(original, new)
def folder_check(folder_path):
i=0
file_list = os.listdir(folder_path)
print(file_list)
duplicate_dict={}
for file in file_list:
# print(file)
file_path=os.path.join(folder_path,file)
for file_compare in file_list:
print(i)
i+=1
file_compare_path=os.path.join(folder_path,file_compare)
if file_compare!=file:
similarity_score=check(file_path,file_compare_path)
# print(str(similarity_score))
if similarity_score==0.0:
print(file,file_compare)
duplicate_dict[file]=file_compare
file_list.remove(str(file))
return duplicate_dict
start_time=time.time()
print(folder_check(r"C:\Users\Admin\Linear-Regression-1\image-similarity-measures\input1"))
end_time=time.time()
stamp=end_time-start_time
print(stamp)

Related

Fast loading multiple .npy files into data generator

I do have thousands of .npy files stored in my hard disk, each containing a single matrix with dimensions [128, T], where T is variable (on average T=800). Each .npy file has size around 2Mb, depending on the matrix shape.
These matrices are then passed to a generator, which yields batches of 32 to a neural network. The Python code used to pass the matrices into the generator is:
def load_batch(path_list):
np_list = []
for path in path_list:
np_list.append(np.load(path))
return np_list
which, given a list of paths of the .npy files, returns a list of the corresponding NumPy matrices.
This code takes, on average, 0.6s to return a list of 32 matrices. I am using append because this is usually a quick operation.
I am aware that the speed of the hard disk buffer does have an influence on timings but, right now, I really would like to shrink the amount of time required as much as possible by just modifying the code in a smart way.
As an alternative, I tried implementing multi-processing:
from multiprocessing import Pool
def reader(filename):
return np.load(filename)
def load_multiprocess(path_list, n_cores=5):
pool = Pool(n_cores)
np_list = pool.map(reader, path_list)
return np_list
However, the performance is much worse. I had a look around stackoverflow, and I got the idea that my specific application could not benefit from multiprocessing.
To summarize, I am looking for any kind of advice for one of these two tasks:
Improving the speed of the first code (even 0.1s less would mean a lot).
Using multiprocessing in the right way, if possible.
SOLUTION AND BENCHMARK
Out of the three methods here proposed, user7138814's solution seems to generally improve a lot the execution speed. However, things seem to change when the data is loaded while training a neural network: even though mapping is by itself still the quicker method for loading data, the overall training time seems to increase, I have no idea where and why, as timings using the mapping load are always better.
Below, I will do a benchmark of the three methods.First, define the methods:
import numpy as np
# my initial method
def load_batch(path_list):
np_list = []
for path in path_list:
np_list.append(np.load(path))
return np_list
# Aaj Kaal's method
def load_batch1(path_list):
return [np.load(path) for path in path_list]
# user7138814's method
def load_batch2(path_list):
np_list = []
for path in path_list:
np_list.append(np.load(path, mmap_mode='r'))
return np_list
I defined a list of paths as follows:
batches_list = []
batch_size = 32
for n in range(0,150):
batches_list.append(X_path_list[n*batch_size:n*batch_size+batch_size])
The list contains 150 batches of 32 paths each, it should be enough to calculate the mean.
Then, each method is executed using passing to it exactly the same data.
import time
# my initial method
timing0 = []
for l in batches_list:
start = time.time()
load_batch(l)
end = time.time()
timing0.append(end-start)
print(np.mean(timing0))
# Aaj Kaal's method
timing1 = []
for l in batches_list:
start = time.time()
load_batch1(l)
end = time.time()
timing1.append(end-start)
print(np.mean(timing1))
# user7138814's method
timing2 = []
for l in batches_list:
start = time.time()
load_batch2(l)
end = time.time()
timing2.append(end-start)
print(np.mean(timing2))
Output (mean timing in seconds over 150 executions):
0.022530150413513184
0.022546884218851725
0.009580903053283692
Results seem to be consistent when changing length of batches_list and batch_size.
Maybe memory mapping the files will be beneficial due to lazy loading. If you would use for example
np.load(filename, mmap_mode='r')
the creation of the numpy array becomes almost a no-op, but later in the pipeline you pay the price. This could provide a speedup if it results in processing the data in parallel with reading from disk.
Did you try using use list comprehension. Replace
def load_batch(path_list):
np_list = []
for path in path_list:
np_list.append(np.load(path))
return np_list
with
def load_batch(path_list):
return [np.load(path) for path in path_list]
In fact you can get rid of the function and directly use list comprehension. If functional call is required use lambda

tensorflow TFRecordWriter killing kernel

I am staring playing with tensorflow. I am facing the following problem. I am trying to run an example to do image recognition based on the Stanford Dog Dataset.I am stuck in the step of converting the image and label in TRFRECORDS files.
In the image dataset folder there are 120 sub-folders, one for each breed (label).
If I run the code below with just on sub-folder in run fine (Actually I didn't tried to read the trfrecord file). But If I include a second sub-folder the process kills the python kernel process.
Here is the code I am running
import glob
import tensorflow as tf
from itertools import groupby
from collections import defaultdict
image_filenames = glob.glob(r'C:\Users\Administrator\Documents\Tensorflow\images\n02*\*.jpg')
training_dataset = defaultdict(list)
testing_dataset = defaultdict(list)
# Split up the filename into its breed and corresponding filename. The breed is found by taking the directo
image_filename_with_breed =map(lambda filename: (filename.split("\\")[6], filename), image_filenames)
# Group each image by the breed which is the 0th element in the tuple returned above
for dog_breed, breed_images in groupby(image_filename_with_breed, lambda x: x[0]):
# Enumerate each breed's image and send ~20% of the images to a testing set
for i, breed_image in enumerate(breed_images):
if i % 5 == 0:
testing_dataset[dog_breed].append(breed_image[1])
else:
training_dataset[dog_breed].append(breed_image[1])
# Check that each breed includes at least 18% of the images for testing
breed_training_count = len(training_dataset[dog_breed])
breed_testing_count = len(testing_dataset[dog_breed])
assert round(breed_testing_count / (breed_training_count + breed_testing_count), 2) > 0.18,'Not enough testing data'
sess = tf.Session()
def write_records_file(dataset, record_location):
"""
Fill a TFRecords file with the images found in `dataset` and include their category.
Parameters
----------
dataset : dict(list)
Dictionary with each key being a label for the list of image filenames of its value.
record_location : str
Location to store the TFRecord output.
"""
writer = None
# Enumerating the dataset because the current index is used to breakup the files if they get over 100
# images to avoid a slowdown in writing.
current_index = 0
for breed, images_filenames in dataset.items():
for image_filename in images_filenames:
print(image_filename)
if current_index % 100 == 0:
if writer:
writer.close()
record_filename = "{record_location}-{current_index}.tfrecords".format(
record_location=record_location,
current_index=current_index)
print(record_filename)
writer = tf.python_io.TFRecordWriter(record_filename)
current_index += 1
image_file = tf.read_file(image_filename)
# In ImageNet dogs, there are a few images which TensorFlow doesn't recognize as JPEGs. This
# try/catch will ignore those images.
try:
image = tf.image.decode_jpeg(image_file)
except:
print(image_filename)
continue
# Converting to grayscale saves processing and memory but isn't required.
grayscale_image = tf.image.rgb_to_grayscale(image)
resized_image = tf.image.resize_images(grayscale_image, (250, 151))
# tf.cast is used here because the resized images are floats but haven't been converted into
# image floats where an RGB value is between [0,1).
image_bytes = sess.run(tf.cast(resized_image, tf.uint8)).tobytes()
# Instead of using the label as a string, it'd be more efficient to turn it into either an
# integer index or a one-hot encoded rank one tensor.
# https://en.wikipedia.org/wiki/One-hot
image_label = breed.encode("utf-8")
example = tf.train.Example(features=tf.train.Features(feature={
'label': tf.train.Feature(bytes_list=tf.train.BytesList(value=[image_label])),
'image': tf.train.Feature(bytes_list=tf.train.BytesList(value=[image_bytes]))
}))
writer.write(example.SerializeToString())
writer.close()
write_records_file(testing_dataset, r'C:\Users\Administrator\Documents\Tensorflow\TRF\testing_images')
write_records_file(training_dataset, r'C:\Users\Administrator\Documents\Tensorflow\TRF\training_images')
I monitored the memory usage and running the script does not seems to consume to much memory. I tried this in two Virtual Machines. One with Ubuntu and the other on with Windows 2000.
Does anyone have a idea?
Thanks!
I found the problem. It was the writer.close() statement that was incorrectly idented. I should be idented in the first for loop but I idented it in the second loop.

Check if files in dir are the same

I have a folder of 5000+ images in jpeg/png etc. How can I check if any of the images are the same. The images were collected through web scraping and have been sequentially renamed so I cannot compare file names.
I am currently checking if the hashes are the same however this is a very long process. I am currently using:
def sameIm(file_name1,file_name2):
hash = imagehash.average_hash(Image.open(path + file_name1))
otherhash = imagehash.average_hash(Image.open(path + file_name2))
return (hash == otherhash)
Then nested loops. Comparing 1 image to 5000+ others takes about 5mins so comparing each to each would take days to compute.
Is there a faster way to do this in python. I was thinking parallel processing but would that still take a long time?
or is there another way to compare files which is faster?
Thanks
There is indeed a much faster way of doing this:
import collections
import glob
import os
def dupDetector(dirpath, ext):
hashes = collections.defaultdict(list)
for fpath in glob.glob(os.path.join(dirpath, "*.{}".format(ext))):
h = imagehash.average_hash(Image.open(fpath))
hashes[h].append(fpath)
for h,fpaths in hashes.items():
if len(fpaths) == 1:
print(fpaths[0], "is one of a kind")
continue
print("The following files are duplicates of each other (with the hash {}): \n\t{}".format(h, '\n\t'.join(fpaths)))
Using the dictionary with the file hash as a key gives you O(1) lookups, which means you don't need to do the pair-wise comparisons. You therefore go from a quadratic runtime, to a linear runtime (yay!)
Why not compute hash only once?
hashes = [imagehash.average_hash(Image.open(path + fn)) for fn in file_names]
def compare_hashes(hash1, hash2):
return hash1 == hash2
One solution is to keep using the hash but stocking it in a list of tuple (or a dic, i don't know wich is more efficient here) where the first element is the name of the image and the second is the hash. It should take aproximatively the same 5 mins.
If you have 5000 images,
You compare the value of the first element of the list to the 4999 others
Then the second to the 4998 others (as you already checked the first one)
Then the third ...
This "just" make you do n²/2 comparisons (where n is the number of images)
Just use map structure to calculate hashes for each image,then store hashes as a key and name of the image as a value.
As a result you would have unique images names array.
def get_hash(filename):
return imagehash.average_hash(Image.open(path + filename))
def get_unique_images(filenames):
hashes = {}
for filename in filenames:
image_hash = get_hash(filename)
hashes[image_hash] = filename
return hashes.values()

Using astropy.fits and numpy to apply coincidence corrections to SWIFT fits image

This question may be a little specialist, but hopefully someone might be able to help. I normally use IDL, but for developing a pipeline I'm looking to use python to improve running times.
My fits file handling setup is as follows:
import numpy as numpy
from astropy.io import fits
#Directory: /Users/UCL_Astronomy/Documents/UCL/PHASG199/M33_UVOT_sum/UVOTIMSUM/M33_sum_epoch1_um2_norm.img
with fits.open('...') as ima_norm_um2:
#Open UVOTIMSUM file once and close it after extracting the relevant values:
ima_norm_um2_hdr = ima_norm_um2[0].header
ima_norm_um2_data = ima_norm_um2[0].data
#Individual dimensions for number of x pixels and number of y pixels:
nxpix_um2_ext1 = ima_norm_um2_hdr['NAXIS1']
nypix_um2_ext1 = ima_norm_um2_hdr['NAXIS2']
#Compute the size of the images (you can also do this manually rather than calling these keywords from the header):
#Call the header and data from the UVOTIMSUM file with the relevant keyword extensions:
corrfact_um2_ext1 = numpy.zeros((ima_norm_um2_hdr['NAXIS2'], ima_norm_um2_hdr['NAXIS1']))
coincorr_um2_ext1 = numpy.zeros((ima_norm_um2_hdr['NAXIS2'], ima_norm_um2_hdr['NAXIS1']))
#Check that the dimensions are all the same:
print(corrfact_um2_ext1.shape)
print(coincorr_um2_ext1.shape)
print(ima_norm_um2_data.shape)
# Make a new image file to save the correction factors:
hdu_corrfact = fits.PrimaryHDU(corrfact_um2_ext1, header=ima_norm_um2_hdr)
fits.HDUList([hdu_corrfact]).writeto('.../M33_sum_epoch1_um2_corrfact.img')
# Make a new image file to save the corrected image to:
hdu_coincorr = fits.PrimaryHDU(coincorr_um2_ext1, header=ima_norm_um2_hdr)
fits.HDUList([hdu_coincorr]).writeto('.../M33_sum_epoch1_um2_coincorr.img')
I'm looking to then apply the following corrections:
# Define the variables from Poole et al. (2008) "Photometric calibration of the Swift ultraviolet/optical telescope":
alpha = 0.9842000
ft = 0.0110329
a1 = 0.0658568
a2 = -0.0907142
a3 = 0.0285951
a4 = 0.0308063
for i in range(nxpix_um2_ext1 - 1): #do begin
for j in range(nypix_um2_ext1 - 1): #do begin
if (numpy.less_equal(i, 4) | numpy.greater_equal(i, nxpix_um2_ext1-4) | numpy.less_equal(j, 4) | numpy.greater_equal(j, nxpix_um2_ext1-4)): #then begin
#UVM2
corrfact_um2_ext1[i,j] == 0
coincorr_um2_ext1[i,j] == 0
else:
xpixmin = i-4
xpixmax = i+4
ypixmin = j-4
ypixmax = j+4
#UVM2
ima_UVM2sum = total(ima_norm_um2[xpixmin:xpixmax,ypixmin:ypixmax])
xvec_UVM2 = ft*ima_UVM2sum
fxvec_UVM2 = 1 + (a1*xvec_UVM2) + (a2*xvec_UVM2*xvec_UVM2) + (a3*xvec_UVM2*xvec_UVM2*xvec_UVM2) + (a4*xvec_UVM2*xvec_UVM2*xvec_UVM2*xvec_UVM2)
Ctheory_UVM2 = - alog(1-(alpha*ima_UVM2sum*ft))/(alpha*ft)
corrfact_um2_ext1[i,j] = Ctheory_UVM2*(fxvec_UVM2/ima_UVM2sum)
coincorr_um2_ext1[i,j] = corrfact_um2_ext1[i,j]*ima_sk_um2[i,j]
The above snippet is where it is messing up, as I have a mixture of IDL syntax and python syntax. I'm just not sure how to convert certain aspects of IDL to python. For example, the ima_UVM2sum = total(ima_norm_um2[xpixmin:xpixmax,ypixmin:ypixmax]) I'm not quite sure how to handle.
I'm also missing the part where it will update the correction factor and coincidence correction image files, I would say. If anyone could have the patience to go over it with a fine tooth comb and suggest the neccessary changes I need that would be excellent.
The original normalised image can be downloaded here: Replace ... in above code with this file
One very important thing about numpy is that it does every mathematical or comparison function on an element-basis. So you probably don't need to loop through the arrays.
So maybe start where you convolve your image with a sum-filter. This can be done for 2D images by astropy.convolution.convolve or scipy.ndimage.filters.uniform_filter
I'm not sure what you want but I think you want a 9x9 sum-filter that would be realized by
from scipy.ndimage.filters import uniform_filter
ima_UVM2sum = uniform_filter(ima_norm_um2_data, size=9)
since you want to discard any pixel that are at the borders (4 pixel) you can simply slice them away:
ima_UVM2sum_valid = ima_UVM2sum[4:-4,4:-4]
This ignores the first and last 4 rows and the first and last 4 columns (last is realized by making the stop value negative)
now you want to calculate the corrections:
xvec_UVM2 = ft*ima_UVM2sum_valid
fxvec_UVM2 = 1 + (a1*xvec_UVM2) + (a2*xvec_UVM2**2) + (a3*xvec_UVM2**3) + (a4*xvec_UVM2**4)
Ctheory_UVM2 = - np.alog(1-(alpha*ima_UVM2sum_valid*ft))/(alpha*ft)
these are all arrays so you still do not need to loop.
But then you want to fill your two images. Be careful because the correction is smaller (we inored the first and last rows/columns) so you have to take the same region in the correction images:
corrfact_um2_ext1[4:-4,4:-4] = Ctheory_UVM2*(fxvec_UVM2/ima_UVM2sum_valid)
coincorr_um2_ext1[4:-4,4:-4] = corrfact_um2_ext1[4:-4,4:-4] *ima_sk_um2
still no loop just using numpys mathematical functions. This means it is much faster (MUCH FASTER!) and does the same.
Maybe I have forgotten some slicing and that would yield a Not broadcastable error if so please report back.
Just a note about your loop: Python's first axis is the second axis in FITS and the second axis is the first FITS axis. So if you need to loop over the axis bear that in mind so you don't end up with IndexErrors or unexpected results.

Create a single-file dataset out of _many_ b/n GIFs [duplicate]

This question already has answers here:
Efficient serialization of numpy boolean arrays
(3 answers)
Closed 8 years ago.
I have many 100x100px black/white GIF images.
I want to use them in Numpy to train a machine learning algorithm, but I would like to save them in a single file that is easily readable in Python/Numpy.
By saying many I mean several hundred thousands, so I would like to take advantage of the images carrying only 1 bit per pixel.
Any idea on how I can do this?
EDIT:
I used a BitArray object, from the bitstring module. Then I saved it using numpy.savez. The problem is that it takes ages to save. I never managed to see the end of such process on the entire dataset. I tried to save a small subset and it took 10 minutes and about 20 times the size of the small subset itself.
I will try with the BoolArray, thanks for the reference.
EDIT (solved):
I solved the problem by using a different approach from those that I found in the questions you linked. I found the numpy.packbits function here: numpy boolean array with 1 bit entries
I'm reporting my code here so it can be useful to others:
accepted_shape = (100, 100)
images = []
for file_path in gifs:
img_data = imread(file_path)
if img_data.shape != accepted_shape:
continue
max_value = img_data.max()
min_value = img_data.min()
middle_value = (max_value - min_value) // 2
image = np.packbits((img_data.ravel() > middle_value).astype(int))
images.append(image)
np.vstack(images)
np.savez_compressed('dataset.npz', shape=accepted_shape, images=images)
This just requires some attention when uncompressing because if the number of bits is not a multiple of 8, some zeros are added as padding. This is how I uncompress the files:
data = np.load('dataset.npz')
shape = data['shape']
images = data['images']
nf = np.prod(shape)
ne = images.size / nf
images = np.unpackbits(images, axis=1)
images = images[:,:nf]
PyTables seems like a good option here. Something like this might work:
import numpy as np
import tables as tb
nfiles = 100000 #or however many files you have
h5file = tb.openFile('data.h5', mode='w', title="Test Array")
root = h5file.root
x = h5file.createCArray(root,'x',tb.Float64Atom(),shape=(100,100,nfiles))
x[:100,:100, 0] = np.random.random(size=(100,100)) # Now put in some data
h5file.close()

Categories

Resources