back to listing index

How-To: Python Compare Two Images - PyImageSearch

[web search]
Original source (
Tags: python image-processing
Clipped on: 2016-06-29

Image (Asset 1/21) alt=

How-To: Python Compare Two Images

By on September 15, 2014 in Image Processing, Tutorials

Image (Asset 2/21) alt=

Would you have guessed that I’m a stamp collector?

Just kidding. I’m not.

But let’s play a little game of pretend.

Let’s pretend that we have a huge dataset of stamp images. And we want to take two arbitrary stamp images and compare them to determine if they are identical, or near identical in some way.

In general, we can accomplish this in two ways.

The first method is to use locality sensitive hashing, which I’ll cover in a later blog post.

The second method is to use algorithms such as Mean Squared Error (MSE) or the Structural Similarity Index (SSIM).

In this blog post I’ll show you how to use Python to compare two images using Mean Squared Error and Structural Similarity Index.

Looking for the source code to this post?
Jump right to the downloads section.

OpenCV and Python versions:
This example will run on Python 2.7/Python 3.4+ and OpenCV 2.4.X/OpenCV 3.0+.

Our Example Dataset

Let’s start off by taking a look at our example dataset:

Image (Asset 3/21) alt=

Figure 1: Our example image dataset. Left: The original image. Middle: The original image with contrast adjustments. Right: The original image with Photoshopped overlay.

Here you can see that we have three images: (left) our original image of our friends from Jurassic Park going on their first (and only) tour, (middle) the original image with contrast adjustments applied to it, and (right), the original image with the Jurassic Park logo overlaid on top of it via Photoshop manipulation.

Now, it’s clear to us that the left and the middle images are more “similar” to each other — the one in the middle is just like the first one, only it is “darker”.

But as we’ll find out, Mean Squared Error will actually say the Photoshopped image is more similar to the original than the middle image with contrast adjustments. Pretty weird, right?

Mean Squared Error vs. Structural Similarity Measure

Let’s take a look at the Mean Squared error equation:

Image (Asset 4/21) alt=

Equation 1: Mean Squared Error

While this equation may look complex, I promise you it’s not.

And to demonstrate this you, I’m going to convert this equation to a Python function:

Python Compare Two Images
def mse(imageA, imageB):
# the 'Mean Squared Error' between the two images is the
# sum of the squared difference between the two images;
# NOTE: the two images must have the same dimension
err = np.sum((imageA.astype("float") - imageB.astype("float")) ** 2)
err /= float(imageA.shape[0] * imageA.shape[1])
# return the MSE, the lower the error, the more "similar"
# the two images are
return err

So there you have it — Mean Squared Error in only four lines of Python code once you take out the comments.

Let’s tear it apart and see what’s going on:

  • On Line 7 we define our mse function, which takes two arguments: imageA and imageB (i.e. the images we want to compare for similarity).
  • All the real work is handled on Line 11. First we convert the images from unsigned 8-bit integers to floating point, that way we don’t run into any problems with modulus operations “wrapping around”. We then take the difference between the images by subtracting the pixel intensities. Next up, we square these difference (hence mean squared error, and finally sum them up.
  • Line 12 handles the mean of the Mean Squared Error. All we are doing is dividing our sum of squares by the total number of pixels in the image.
  • Finally, we return our MSE to the caller one Line 16.

MSE is dead simple to implement — but when using it for similarity, we can run into problems. The main one being that large distances between pixel intensities do not necessarily mean the contents of the images are dramatically different. I’ll provide some proof for that statement later in this post, but in the meantime, take my word for it.

It’s important to note that a value of 0 for MSE indicates perfect similarity. A value greater than one implies less similarity and will continue to grow as the average difference between pixel intensities increases as well.

In order to remedy some of the issues associated with MSE for image comparison, we have the Structural Similarity Index, developed by Wang et al.:

Image (Asset 5/21) alt=

Equation 2: Structural Similarity Index

The SSIM method is clearly more involved than the MSE method, but the gist is that SSIM attempts to model the perceived change in the structural information of the image, whereas MSE is actually estimating the perceived errors. There is a subtle difference between the two, but the results are dramatic.

Furthermore, the equation in Equation 2 is used to compare two windows (i.e. small sub-samples) rather than the entire image as in MSE. Doing this leads to a more robust approach that is able to account for changes in the structure of the image, rather than just the perceived change.

The parameters to Equation 2 include the (x, y) location of the N x N window in each image, the mean of the pixel intensities in the x and y direction, the variance of intensities in the x and y direction, along with the covariance.

Unlike MSE, the SSIM value can vary between -1 and 1, where 1 indicates perfect similarity.

Luckily, as you’ll see, we don’t have to implement this method by hand since scikit-image already has an implementation ready for us.

Let’s go ahead and jump into some code.

How-To: Compare Two Images Using Python

Python Compare Two Images
# import the necessary packages
from skimage.measure import structural_similarity as ssim
import matplotlib.pyplot as plt
import numpy as np
import cv2

We start by importing the packages we’ll need — matplotlib for plotting, NumPy for numerical processing, and cv2 for our OpenCV bindings. Our Structural Similarity Index method is already implemented for us by scikit-image, so we’ll just use their implementation.

Python Compare Two Images
def mse(imageA, imageB):
# the 'Mean Squared Error' between the two images is the
# sum of the squared difference between the two images;
# NOTE: the two images must have the same dimension
err = np.sum((imageA.astype("float") - imageB.astype("float")) ** 2)
err /= float(imageA.shape[0] * imageA.shape[1])
# return the MSE, the lower the error, the more "similar"
# the two images are
return err
def compare_images(imageA, imageB, title):
# compute the mean squared error and structural similarity
# index for the images
m = mse(imageA, imageB)
s = ssim(imageA, imageB)
# setup the figure
fig = plt.figure(title)
plt.suptitle("MSE: %.2f, SSIM: %.2f" % (m, s))
# show first image
ax = fig.add_subplot(1, 2, 1)
plt.imshow(imageA, cmap =
# show the second image
ax = fig.add_subplot(1, 2, 2)
plt.imshow(imageB, cmap =
# show the images

Lines 7-16 define our mse method, which you are already familiar with.

We then define the compare_images function on Line 18 which we’ll use to compare two images using both MSE and SSIM. The mse function takes three arguments: imageA and imageB, which are the two images we are going to compare, and then the title of our figure.

We then compute the MSE and SSIM between the two images on Lines 21 and 22.

Lines 25-39 handle some simple matplotlib plotting. We simply display the MSE and SSIM associated with the two images we are comparing.

Python Compare Two Images
# load the images -- the original, the original + contrast,
# and the original + photoshop
original = cv2.imread("images/jp_gates_original.png")
contrast = cv2.imread("images/jp_gates_contrast.png")
shopped = cv2.imread("images/jp_gates_photoshopped.png")
# convert the images to grayscale
original = cv2.cvtColor(original, cv2.COLOR_BGR2GRAY)
contrast = cv2.cvtColor(contrast, cv2.COLOR_BGR2GRAY)
shopped = cv2.cvtColor(shopped, cv2.COLOR_BGR2GRAY)

Lines 43-45 handle loading our images off disk using OpenCV. We’ll be using our original image (Line 43), our contrast adjusted image (Line 44), and our Photoshopped image with the Jurassic Park logo overlaid (Line 45).

We then convert our images to grayscale on Lines 48-50.

Python Compare Two Images
# initialize the figure
fig = plt.figure("Images")
images = ("Original", original), ("Contrast", contrast), ("Photoshopped", shopped)
# loop over the images
for (i, (name, image)) in enumerate(images):
# show the image
ax = fig.add_subplot(1, 3, i + 1)
plt.imshow(image, cmap =
# show the figure
# compare the images
compare_images(original, original, "Original vs. Original")
compare_images(original, contrast, "Original vs. Contrast")
compare_images(original, shopped, "Original vs. Photoshopped")

Now that our images are loaded off disk, let’s show them. On Lines 52-65 we simply generate a matplotlib figure, loop over our images one-by-one, and add them to our plot. Our plot is then displayed to us on Line 65.

Finally, we can compare our images together using the compare_images function on Lines 68-70.

We can execute our script by issuing the following command:

Python Compare Two Images
$ python


Once our script has executed, we should first see our test case — comparing the original image to itself:

Image (Asset 6/21) alt=

Figure 2: Comparing the two original images together.

Not surpassingly, the original image is identical to itself, with a value of 0.0 for MSE and 1.0 for SSIM. Remember, as the MSE increases the images are less similar, as opposed to the SSIM where smaller values indicate less similarity.

Now, take a look at comparing the original to the contrast adjusted image:

Image (Asset 7/21) alt=

Figure 3: Comparing the original and the contrast adjusted image.

In this case, the MSE has increased and the SSIM decreased, implying that the images are less similar. This is indeed true — adjusting the contrast has definitely “damaged” the representation of the image.

But things don’t get interesting until we compare the original image to the Photoshopped overlay:

Image (Asset 8/21) alt=

Figure 4: Comparing the original and Photoshopped overlay image.

Comparing the original image to the Photoshop overlay yields a MSE of 1076 and a SSIM of 0.69.

Wait a second.

A MSE of 1076 is smaller than the previous of 1401. But clearly the Photoshopped overlay is dramatically more different than simply adjusting the contrast! But again, this is a limitation we must accept when utilizing raw pixel intensities globally.

On the other end, SSIM is returns a value of 0.69, which is indeed less than the 0.78 obtained when comparing the original image to the contrast adjusted image.


In this blog post I showed you how to compare two images using Python.

To perform our comparison, we made use of the Mean Squared Error (MSE) and the Structural Similarity Index (SSIM) functions.

While the MSE is substantially faster to compute, it has the major drawback of (1) being applied globally and (2) only estimating the perceived errors of the image.

On the other hand, SSIM, while slower, is able to perceive the change in structural information of the image by comparing local regions of the image instead of globally.

So which method should you use?

It depends.

In general, SSIM will give you better results, but you’ll lose a bit of performance.

But in my opinion, the gain in accuracy is well worth it.

Definitely give both MSE and SSIM a shot and see for yourself!


Image (Asset 9/21) alt=If you would like to download the code and images used in this post, please enter your email address in the form below. Not only will you get a .zip of the code, I’ll also send you a FREE 11-page Resource Guide on Computer Vision and Image Search Engines, including exclusive techniques that I don’t post on this blog! Sound good? If so, enter your email address and I’ll send you the code immediately!

Email address:

Resource Guide (it’s totally free).

Image (Asset 10/21) alt=

Enter your email address below to get my free 11-page Image Search Engine Resource Guide PDF. Uncover exclusive techniques that I don't publish on this blog and start building image search engines of your own!

45 Responses to How-To: Python Compare Two Images

  1. Image (Asset 11/21) alt=
    Xavier Paul November 26, 2014 at 4:53 am #

    Good day Adrian, I am trying to do a program that will search for an Image B within an Image A. I’m able to do with C# but it takes about 6seconds to detect image B in A and report its coordinates.

    Please can you help me?

    • Image (Asset 12/21) alt=
      Adrian Rosebrock November 26, 2014 at 7:14 am #

      Hi Xavier. I think you might want to take a look at template matching to do this. I did a guest post over at Machine Learning Mastery on how to do this.

  2. Image (Asset 13/21) alt=
    Mark December 4, 2014 at 11:33 pm #

    Marvellous! It’s in a very good way to describe and teach. Thanks for the great work.

    Next step, would it be possible to mark the difference between the 2 pictures?

    Below is a simple way, but I am much looking forward to see an advance one. Thanks.

    from PIL import Image
    from PIL import ImageChops
    from PIL import ImageDraw


    dif = ImageChops.difference(imageB, imageA).getbbox()
    draw = ImageDraw.Draw(imageB)

    • Image (Asset 14/21) alt=
      Adrian Rosebrock December 5, 2014 at 7:15 am #

      Hi Mark, a very simple way to visualize the difference between two images is to simply subtract one from the other. OpenCV has a built in function for this called cv2.subtract.

  3. Image (Asset 15/21) alt=
    Mark December 9, 2014 at 2:10 am #

    Hello dear Doctor Rosebrock,

    Many thanks for your reply, and guidance.

    I googled and I can only find some examples involved cv2.subtract for other purposes but not marking differences between 2 pictures.

    You have November and December posts using cv2.subtract but newbie like me just don’t get how to make it work to mark difference.

    Would you mind give an example, if you have time? Thanks.

    • Image (Asset 16/21) alt=
      Adrian Rosebrock December 9, 2014 at 7:30 am #

      Hi Mark, if I understand correctly, are you trying to visualize the difference between two afters after applying the cv2.subtract function? If so, all you need to do is apply cv2.imshow on the output of cv2.subtract. Then you’ll be able to see your output.

  4. Image (Asset 17/21) alt=
    Mridula February 17, 2015 at 12:47 am #

    Hi Adrian,

    I need to compare 2 images under a masked region. Can you help me with that? I mean how to i extend this code to work for a subregion of the images. Also i do a lot of video processing, like comparing whether 2 videos are equal or whether the videos have any artifacts. I would like to make it automated. Any posts on that?

    • Image (Asset 18/21) alt=
      Adrian Rosebrock February 17, 2015 at 6:52 am #

      Hi Mridula. If you’re comparing 2 masked regions, you’re probably better off using a feature based approach by extracting features from the 2 masked regions. However, if your 2 masked regions have the same dimensions or aspect ratios, you might be able to get away with SSIM or MSE. And if you’re interested in comparing two images/frames to see if they are identical, I would utilize image hashing.

    • Image (Asset 19/21) alt=
      Umesh February 22, 2016 at 5:34 pm #

      Hi Mridula,
      I am looking something similar to what you are doing on automation of comparing 2 videos..
      Any input on what you are using and go ahead..
      Thanks in advance

  5. Image (Asset 20/21) alt=
    bhavesh March 19, 2015 at 2:51 am #

    can you guide regarding how to compare two card , one image (card) is stored in disk and second image(card ) to be compare has been taken from camera

    • Image (Asset 21/21) alt=
      Adrian Rosebrock March 19, 2015 at 7:05 am #

      Hi Bhavesh, if you are looking to capture an image from your webcam, take a look a this post to get you started. It shows an example on how to access your webcam using Python + OpenCV. From there, you can take the code and modify it to your needs!

  6. Weston Renou March 19, 2015 at 4:13 pm #

    Thanks for this. I’ve inadvertently duplicated some of my personal photos and I wanted a quick way to de-duplicate my photos *and* a good entry project to start playing with computer vision concepts and techniques. And this is a perfect little project.

  7. Hi,

    I am trying to evaluate the segmentation performance between segmented image and ground truth in binary image. In this case, which metric is suitable to compare?

    Thank you.

    • That’s a great question. In reality, there are a lot of different methods that you could use to evaluate your segmentation. However, I would let your overall choice be defined by what others are using in the literature. Personally, I have not had to evaluate segmentation algorithms analytically before, so I would start by reading segmentation survey papers such as these and seeing what metrics others are using.

  8. How do I compare images of different sizes?

  9. Hi Adrian,

    That was a very informative post and well explained. I have it working with png images, do you know if it’s possible to compare dicom images using the same method?

    I have tried using the pydicom package but have not had any success.

    Any help or advice would be greatly appreciated!

  10. Thank you for this great post. I am wondering how post about locality sensitive hashing is advancing?

    • Hey Primoz, thanks for the comment. Locality Sensitive Hashing is a great topic, I’ll add it to my queue of ideas to write about. My favorite LSH methods use random projections to construct randomly separating hyperplanes. Then, for each hyperplane, a hash is constructed based on which “side” the feature lies on. These binary tests can be compiled into a hash. It’s a neat, super efficient trick that tends to perform well in the real-world. I’ll be sure to do a post about it in the future!

  11. Thank you for this great post. I am working on it . I would like to know how to convert the MSE to the percentage difference of the two images.

  12. budy August 9, 2015 at 10:37 pm #

    nice explanation….thanks

  13. Hi Adrian
    Is there a way or a method exposed by scikit-image to write the diff between two images used to compare to another file?
    Also,Is there any way to ignore some elements/text in image to be compared?

    • To write the difference between two images to file, you could just use normal subtraction and subtract the two images from each other, followed by writing them to file. As for ignoring certain elements in the image, no, that cannot be done without heavily modifying the SSIM or MSE function.

  14. Hi Adrian, I have tried a lot to install skimage library for python 2.7. but it seems there is a problem with the installations. am not able to get any help. is there anyother possible package that could help regarding the same? I am actually trying to implement GLCM and Haralick features for finding out texture parameters. Also, is there any other site that can help regarding the Skimage library problem??

  15. Hi Adrian,
    I am working on a project in which i need to compare the two videos and give an out put with the difference between the reference video and the actual converted video. And this white process needs to be automated..
    Any input on this.
    Thanks in advance

  16. hi adrian…..I am working on a project in which i need to compare the image already stored with the person in the live streaming and i want to check whether those persons are same.
    Thanks in advance

  17. I want to compare an object captured from live streaming video with already stored image of the object.But i cant find the correct code for executing this.please help me

  18. Hi Adrian, read your article and is quite helpful in what I am trying to achieve. Actually I am implementing algorithm for converting grayscale image to colored one based on the given grayscale image and an almost similar colored image. I have implemented it and now want to see how close is the resulting image to the given colored image. I have gone through your article and implemented what you have given here.
    1. Is there any other method to do so for colored images or will the same methods (MSE, SSIM and Locality Sensitive Hashing) work fine?
    2. Also, I read the paper related to SSIM in which it was written that SSIM works for grayscale images only. Is it really so?

    • SSIM is normally only applied to a single channel at a time. Traditionally, this normally means grayscale images. However, in both the case of MSE and SSIM just split the image into its respective Red, Green, and Blue channels, apply the metric, and then take the sum the errors/accuracy. This can be a proxy accuracy for the colorization of your image.

  19. How can I compare stored image and capturing image as per the pixel to pixel comparison for open CV python for the Raspberry Pi

  20. Hi,
    Very useful article, for a beginner like me.
    I want to compare two “JPG” images captured pi cam, and in result give a bit to GPIO
    images are stored in Pi SD card.
    please help

    • There are various algorithms you can use to compare two images. I’ve detailed MSE and SSIM in this blog post. You could also compare images based on their color (histograms, moments), texture (LBPs, textons, Haralick), or even shape (Hu moments, Zernike moments). There is also keypoint matching methods which I discuss inside Practical Python and OpenCV. As for passing the result bit to the GPIO, be sure to read this blog post where I demonstrate how to use GPIO + OpenCV together. Next week’s blog post will also discuss this same topic.

  21. Hi Adrian,

    I am working in photgrammetry and 3D reconstruction.When the user clicks a point in the first image,i want that point to be automatically to be detected in the second image without the user selecting the point in the second image as it leads to large errors.How can this be done,i have tried cropping the portion around the point and trying to match it through brute force matcher and ORB.However it detects no points.
    Please suggest a technique!!
    I can solve for the point mathematically but i want to use image processing to get the point.

Leave a Reply

Name (required)

Email (will not be published) (required)


Resource Guide (it’s totally free).

Click the button below to get my free 11-page Image Search Engine Resource Guide PDF. Uncover exclusive techniques that I don't publish on this blog and start building image search engines of your own.

You can detect faces in images & video.

Are you interested in detecting faces in images & video? But tired of Googling for tutorials that never work? Then let me help! I guarantee that my new book will turn you into a face detection ninja by the end of this weekend. Click here to give it a shot yourself.

PyImageSearch Gurus: NOW ENROLLING!

The PyImageSearch Gurus course is now enrolling! Inside the course you'll learn how to perform:

  • Automatic License Plate Recognition (ANPR)
  • Deep Learning
  • Face Recognition
  • and much more!

Click the button below to learn more about the course, take a tour, and get 10 (FREE) sample lessons.

Hello! I’m Adrian Rosebrock.

I'm an entrepreneur and Ph.D who has launched two successful image search engines, ID My Pill and Chic Engine. I'm here to share my tips, tricks, and hacks I've learned along the way.

Learn computer vision in a single weekend.

Want to learn computer vision & OpenCV? I can teach you in a single weekend. I know. It sounds crazy, but it’s no joke. My new book is your guaranteed, quick-start guide to becoming an OpenCV Ninja. So why not give it a try? Click here to become a computer vision ninja.

Subscribe via RSS

Never miss a post! Subscribe to the PyImageSearch RSS Feed and keep up to date with my image search engine tutorials, tips, and tricks

Find me on Twitter, Facebook, Google+, and LinkedIn.

© 2016 PyImageSearch. All Rights Reserved.


Free 21-day crash course on computer vision & image search engines

Interested in computer vision and image search engines, but don't know where to start? Let me help. I've created a free, 21-day crash course that is hand-tailored to give you the best possible introduction to computer vision. Sound good? Enter your email below to start your journey to becoming a computer vision master.