Detecting cats in images with OpenCV

Did you know that OpenCV can detect cat faces in images…right out-of-the-box with no extras?

I didn’t either.

But after Kendrick Tan broke the story, I had to check it out for myself…and do a little investigative work to see how this cat detector seemed to sneak its way into the OpenCV repository without me noticing (much like a cat sliding into an empty cereal box, just waiting to be discovered).

In the remainder of this blog post, I’ll demonstrate how to use OpenCV’s cat detector to detect cat faces in images. This same technique can be applied to video streams as well.

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

Detecting cats in images with OpenCV

If you take a look at the OpenCV repository, specifically within the haarcascades directory (where OpenCV stores all its pre-trained Haar classifiers to detect various objects, body parts, etc.), you’ll notice two files:

  • haarcascade_frontalcatface.xml
  • haarcascade_frontalcatface_extended.xml

Both of these Haar cascades can be used detecting “cat faces” in images. In fact, I used these very same cascades to generate the example image at the top of this blog post.

Doing a little investigative work, I found that the cascades were trained and contributed to the OpenCV repository by the legendary Joseph Howse who’s authored a good many tutorials, books, and talks on computer vision.

In the remainder of this blog post, I’ll show you how to utilize Howse’s Haar cascades to detect cats in images.

Cat detection code

Let’s get started detecting cats in images with OpenCV. Open up a new file, name it , and insert the following code:

Lines 2 and 3 import our necessary Python packages while Lines 6-12 parse our command line arguments. We only require a single argument here, the input --image  that we want to detect cat faces in using OpenCV.

We can also (optionally) supply a path our Haar cascade via the --cascade  switch. We’ll default this path to haarcascade_frontalcatface.xml  and assume you have the haarcascade_frontalcatface.xml  file in the same directory as your  script.

Note: I’ve conveniently included the code, cat detector Haar cascade, and example images used in this tutorial in the “Downloads” section of this blog post. If you’re new to working with Python + OpenCV (or Haar cascades), I would suggest downloading the provided .zip file to make it easier to follow along.

Next, let’s detect the cats in our input image:

On Lines 15 and 16 we load our input image from disk and convert it to grayscale (a normal pre-processing step before passing the image to a Haar cascade classifier, although not strictly required).

Line 20 loads our Haar cascade from disk (in this case, the cat detector) and instantiates the cv2.CascadeClassifier  object.

Detecting cat faces in images with OpenCV is accomplished on Lines 21 and 22 by calling the detectMultiScale  method of the detector  object. We pass four parameters to the detectMultiScale  method, including:

  1. Our image, gray , that we want to detect cat faces in.
  2. A scaleFactor  of our image pyramid used when detecting cat faces. A larger scale factor will increase the speed of the detector, but could harm our true-positive detection accuracy. Conversely, a smaller scale will slow down the detection process, but increase true-positive detections. However, this smaller scale can also increase the false-positive detection rate as well. See the “A note on Haar cascades” section of this blog post for more information.
  3. The minNeighbors  parameter controls the minimum number of detected bounding boxes in a given area for the region to be considered a “cat face”. This parameter is very helpful in pruning false-positive detections.
  4. Finally, the minSize  parameter is pretty self-explanatory. This value ensures that each detected bounding box is at least width x height pixels (in this case, 75 x 75).

The detectMultiScale  function returns rects , a list of 4-tuples. These tuples contain the (x, y)-coordinates and width and height of each detected cat face.

Finally, let’s draw a rectangle surround each cat face in the image:

Given our bounding boxes (i.e., rects ), we loop over each of them individually on Line 25.

We then draw a rectangle surrounding each cat face on Line 26, while Lines 27 and 28 displays an integer, counting the number of cats in the image.

Finally, Lines 31 and 32 display the output image to our screen.

Cat detection results

To test our OpenCV cat detector, be sure to download the source code to this tutorial using the “Downloads” section at the bottom of this post.

Then, after you have unzipped the archive, you should have the following three files/directories:

  1. : Our Python + OpenCV script used to detect cats in images.
  2. haarcascade_frontalcatface.xml : The cat detector Haar cascade.
  3. images : A directory of testing images that we’re going to apply the cat detector cascade to.

From there, execute the following command:

Figure 1: Detecting a cat face in an image, even with parts of the cat occluded.

Figure 1: Detecting a cat face in an image, even with parts of the cat occluded (source).

Notice that we have been able to detect the cat face in the image, even though the rest of its body is obscured.

Let’s try another image:

Figure 2: A second example of detecting a cat in an image with OpenCV, this time the cat face is slightly different.

Figure 2: A second example of detecting a cat in an image with OpenCV, this time the cat face is slightly different (source).

This cat’s face is clearly different from the other one, as it’s in the middle of a “meow”. In either case, the cat detector cascade is able to correctly find the cat face in the image.

The same is true for this image as well:

Figure 3: Cat detection with OpenCV and Python (source)

Figure 3: Cat detection with OpenCV and Python (source).

Our final example demonstrates detecting multiple cats in an image using OpenCV and Python:

Figure 4: Detecting multiple cats in the same image with OpenCV (source).

Figure 4: Detecting multiple cats in the same image with OpenCV (source).

Note that the Haar cascade can return bounding boxes in an order that you may not like. In this case, the middle cat is actually labeled as the third cat. You can resolve this “issue” by sorting the bounding boxes according to their (x, y)-coordinates for a consistent ordering.

A quick note on accuracy

It’s important to note that in the comments section of the .xml  files, Joseph Howe details that the cat detector Haar cascades can report cat faces where there are actually human faces.

In this case, he recommends performing both face detection and cat detection, then discarding any cat bounding boxes that overlap with the face bounding boxes.

A note on Haar cascades

First published in 2001 by Paul Viola and Michael Jones, Rapid Object Detection using a Boosted Cascade of Simple Features, this original work has become one of the most cited papers in computer vision.

This algorithm is capable of detecting objects in images, regardless of their location and scale. And perhaps most intriguing, the detector can run in real-time on modern hardware.

In their paper, Viola and Jones focused on training a face detector; however, the framework can also be used to train detectors for arbitrary “objects”, such as cars, bananas, road signs, etc.

The problem?

The biggest problem with Haar cascades is getting the detectMultiScale  parameters right, specifically scaleFactor  and minNeighbors . You can easily run into situations where you need to tune both of these parameters on an image-by-image basis, which is far from ideal when utilizing an object detector.

The scaleFactor  variable controls your image pyramid used to detect objects at various scales of an image. If your scaleFactor  is too large, then you’ll only evaluate a few layers of the image pyramid, potentially leading to you missing objects at scales that fall in between the pyramid layers.

On the other hand, if you set scaleFactor  too low, then you evaluate many pyramid layers. This will help you detect more objects in your image, but it (1) makes the detection process slower and (2) substantially increases the false-positive detection rate, something that Haar cascades are known for.

To remember this, we often apply Histogram of Oriented Gradients + Linear SVM detection instead.

The HOG + Linear SVM framework parameters are normally much easier to tune — and best of all, HOG + Linear SVM enjoys a much smaller false-positive detection rate. The only downside is that it’s harder to get HOG + Linear SVM to run in real-time.

Interested in learning more about object detection?

Figure 5: Learn how to build custom object detectors inside the PyImageSearch Gurus course.

Figure 5: Learn how to build custom object detectors inside the PyImageSearch Gurus course.

If you’re interested in learning how to train your own custom object detectors, be sure to take a look at the PyImageSearch Gurus course.

Inside the course, I have 15 lessons covering 168 pages of tutorials dedicated to teaching you how to build custom object detectors from scratch.  You’ll discover how to detect road signsfacescars (and nearly any other object) in images by applying the HOG + Linear SVM framework for object detection.

To learn more about the PyImageSearch Gurus course (and grab 10 FREE sample lessons), just click the button below:

Click here to learn more about PyImageSearch Gurus!


In this blog post, we learned how to detect cats in images using the default Haar cascades shipped with OpenCV. These Haar cascades were trained and contributed to the OpenCV project by Joseph Howse, and were originally brought to my attention in this post by Kendrick Tan.

While Haar cascades are quite useful, we often use HOG + Linear SVM instead, as it’s a bit easier to tune the detector parameters, and more importantly, we can enjoy a much lower false-positive detection rate.

I detail how to build custom HOG + Linear SVM object detectors to recognize various objects in images, including cars, road signs, and much more inside the PyImageSearch Gurus course.

Anyway, I hope you enjoyed this blog post!

Before you go, be sure to signup for the PyImageSearch Newsletter using the form below to be notified when new blog posts are published.


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!

, ,

41 Responses to Detecting cats in images with OpenCV

  1. Joseph Howse June 20, 2016 at 6:36 pm #

    Thanks so much for featuring the cat face cascades! I am delighted to see them in action here alongside your other great tutorials!

    Fun facts:

    The haarcascade_frontalcatface_extended.xml version uses the “extended” Haar feature set to make it sensitive to diagonal features such as ears and whiskers. >^-^<

    Besides the Haar versions, you will find an LBP version, lbpcascade_frontalcatface.xml, in OpenCV's lbpcascades folder. The LBP version is less accurate but faster; you might like it for Raspberry Pi or other platforms with limited resources.

    Details about the training of the cat face cascades can be found in my book, OpenCV for Secret Agents, and in free presentations on my website's OpenCV landing page (

    Nine lives,


    • Adrian Rosebrock June 23, 2016 at 1:34 pm #

      You’re the one we should be thanking Joe — you trained the actual cascades! 🙂

  2. Kendrick Tan June 20, 2016 at 11:46 pm #

    Hi Adrian, I’m a bit stoked when I saw my name come up on the newsletter.

    Anyway, I just came here to say thank you. Your tutorials really gave me a firm understanding of the basics of computer vision, and machine learning. Really appreciate your time and effort.

    • Adrian Rosebrock June 23, 2016 at 1:33 pm #

      Hey Kendrick! Thanks for doing the initial awesome post — without your post, the cat detection cascade would have totally passed me by. I should be the one thanking you 🙂

  3. Linus June 21, 2016 at 7:39 am #

    Thanks Adrian, this is so awesome! I’ve thought about the possibitity of detecting animals some time ago, but I’ve never found a solution 😛
    What about other animals (dogs etc.)? Are there also xml files available? And is using this in a live camera feed also possibe, with a moderate execution speed?

    • Adrian Rosebrock June 23, 2016 at 1:29 pm #

      As far as I know, there is only the Haar cascade for cats — there isn’t one for dogs. Although you could certainly train one. I would suggest using HOG + Linear SVM instead since it has a lower false-positive rate.

      And yes, this method can be used to run in real-time. Take a look at Practical Python and OpenCV for more details on running Haar cascades in real-time.

  4. Ranjeet Singh June 21, 2016 at 12:06 pm #

    But now I want to know what that xml file is ? How it has been written ?

    • Adrian Rosebrock June 23, 2016 at 1:27 pm #

      The XML file is a serialized Haar cascade. It is generated by training a machine learning algorithm to recognize various objects in images. To recognize your own objects, you’ll need to train your own custom object detector.

  5. swight November 10, 2016 at 7:23 pm #

    How would I write the final image to file instead of displaying to screen?

    • Adrian Rosebrock November 14, 2016 at 12:16 pm #

      You can just use the cv2.imwrite function:

      cv2.imwrite("/path/to/my/image.jpg", image)

      If you need help learning the basics of computer vision and image processing I would suggest you work through Practical Python and OpenCV.

  6. Victor December 17, 2016 at 9:34 pm #

    We trained an LBP cascade with better time and accuracy performance. If anyone wants this, we can push it to our 10imaging/opencv repo on GitHub.

    • Ridge December 19, 2016 at 8:01 pm #

      @Victor – Sure, I’d like to see it.

  7. Gismo Scha March 15, 2017 at 8:28 pm #

    Thanks Adrian, it works!
    How can i extract / copy the rectangled area and write to file?

    • Adrian Rosebrock March 17, 2017 at 9:31 am #

      You would use NumPy array slicing to extract the bounding box and then use cv2.imwrite to write the image to disk. I discuss these basic operations inside Practical Python and OpenCV. I would suggest you start there if you are new to OpenCV.

  8. Harman Singh March 18, 2017 at 5:06 pm #

    Thank Adrian, your tutorials are great and fun and also detects fairly good but while deploying this script on multiple images with different size, the output is not to accurate.

    for example in 1st pic (cat rolled in paper) i got 2 rectangle one covering the face and other on the nose..(to get 1 rect, i adjusted scaleFactor to 1.715)

    even after converting all the image to size still getting mixed result like mentioned above and sometimes none.

    do i have to adjust scalefactor everytime? or is it me having this problem?

    Thank You.

  9. Videsh Suman May 15, 2017 at 1:44 pm #

    Hello, Adrain! Your blogs are great!
    I recently installed OpenCV on my Mac along with Python 2.7 following your other blog, and tried running this program without the virtual environment. It successfully ran. Why is that so? If my OpenCV was essentially compiled on the virtual environment, it should only be able to run on the cv environment, right?
    Could you please explain?

    • Adrian Rosebrock May 17, 2017 at 10:07 am #

      OpenCV is installed globally on your machine. A Python virtual environment is then created that is separate from the rest of the packages on the system (when we then sym-link the file into). We do this just in case we need to install different versions of NumPy, SciPy, etc. than are already installed on your system.

  10. Kyle Adkins May 29, 2017 at 11:47 pm #

    Hey, Adrian.

    I installed OpenCV on MacOS Sierra using the updated tutorial you made that solves the QTKit problem. But when I tried this code (& whenever I try to use cv2.imshow) I get a blank window.

    The name of the window shows up, and so does the yellow Mac minimize button. But that’s it, no image. And I can’t exit the window unless I force quit the terminal session. I’ve searched Google & StackOverflow but haven’t found any answers. Any idea what’s going on?

    Thanks in advance!

    • Adrian Rosebrock May 31, 2017 at 1:16 pm #

      Hey Kyle — that is super strange. Can you do a print(image.dtype). I’d be curious to see what data type the image is.

      • Jon Mittelbronn June 5, 2017 at 11:03 am #

        Same issue. I get uint8, when I do a print(image.dtype). Thanks

        • Adrian Rosebrock June 6, 2017 at 12:02 pm #

          The uint8 data type is correct. What about print(image.shape). What are the resulting dimensions?

          • Luis July 5, 2017 at 1:50 pm #

            for: –image images/cat_02.jpg
            i got the dimensions: (375, 500, 3)
            But i am too having this problem.

          • Luis July 5, 2017 at 1:53 pm #

            so what i tried:
            i put “print(“random text”)” between chunks of code, and I’m not sure if its suppose to stay in the for loop of “# loop over the cat faces and draw a rectangle surrounding each”. it does not print the test code under this loop. idk if that helps 😛

    • Yichen June 17, 2017 at 10:34 am #

      Hi Kyle and Adrian

      I’m also getting the same error on my cv2.imshow function, it’s only showing a top border of a regular Mac window, with a yellow minimize button on it. However, I can exit the window with any keyboard input. This also happens in the example codes from your book. I posted this question on StackOverflow:

      And a few of us believe this might be the problem with the latest OpenCV version (Tested on 17 June 2017). I really hope you could help us through this!
      Thanks in advance!

      • Adrian Rosebrock June 20, 2017 at 11:13 am #

        Hi Yichen — thanks for picking up a copy of Practical Python and OpenCV!

        And thank you for passing along the link to the StackOverflow thread. This definitely sounds like an issue with OpenCV not being able to access the GUI libraries correctly. Were you using the OpenCV 3.2 release or a development release?

        • Luis July 5, 2017 at 1:46 pm #

          yes, i believe he did install 3.2 because that is the one i installed and i too have that same problem 🙁

  11. Danlin Zhang June 16, 2017 at 4:28 pm #

    Hi Adrian,

    I’ve been following your posts for some time and they have been amazing! But for facial-detection, it seems that I cannot get python to display the right result (as it does not display anything, neither the original image nor the detector. I’m running similar issues in the post about facial landmarks with dlib as well. I have installed OpenCV, Homebrew, etc. just as your former posts instructed. Could you help me with that? Thank you very much!

    • Danlin Zhang June 16, 2017 at 4:30 pm #

      I just came across the former post and mine is exactly the same case. I’m also using macOS Sierra and it gives a blank window. I could not exit the window without force-quitting terminal.

      • Adrian Rosebrock June 20, 2017 at 11:17 am #

        Take a look at the comment by “Yichen” below. The issue is that you likely installed a (currently buggy) development version of OpenCV rather than an official release. Try re-compiling and re-installing OpenCV with the official 3.2 release to see if it resolves the issue.

  12. Shai August 19, 2017 at 5:09 pm #

    Thank you for the tutorial!

    I’ve tried running it on both MacOS and Ubuntu 14.04 without success.
    While there are no errors, no cats are identified.

    The length of the tuple of rects is zero :\

    I tried downloading the latest haarcascade file to the folder of the script and giving it the file as an argument, but the result is the same.

    How does one begin to debug this?


    • Adrian Rosebrock August 21, 2017 at 3:42 pm #

      Hi Shai — which version of OpenCV are you using? Did you change any of the parameters from the ones I used in the blog post?

  13. Quan September 11, 2017 at 12:34 pm #

    Thank you for the tutorial!
    I can run it correctly when I use your haarcascade. But I run it with my opencv haarcascade without success. The length of the tuple if rects is zero;

    How can I fix it?


    • Adrian Rosebrock September 11, 2017 at 2:27 pm #

      Haar cascades can be a real pain to tune. You’ll want to adjust the parameters to detectMultiScale. I cover how to tune them and play around with different values inside Practical Python and OpenCV.

  14. manijay September 18, 2017 at 3:43 am #


    I was attempting to run the initial run through with cat_01.jpg, but I encountered a Keyerror on “cascade” as the parameter to CascadeClassifier on line 20. Any solutions?


    • Adrian Rosebrock September 18, 2017 at 2:00 pm #

      What is the exact error message?

  15. manijay September 20, 2017 at 1:59 am #

    In [35]: ls haarcascade_frontalcatface.xml images/

    In [36]: python –image images/cat_02.jpg
    File “”, line 1
    python –image images/cat_02.jpg
    SyntaxError: invalid syntax

    • Adrian Rosebrock September 20, 2017 at 6:58 am #

      You need to download the source code using the “Downloads” section of this post and then execute it via a command line. Do not execute it from with a Python shell/notebook.

  16. Avi September 22, 2017 at 9:29 pm #


    I installed OpenCV with Python following a previous blog using Homebrew.

    When I run this example code on the command line, however, nothing happens. There’s no output window and the process does not terminate. I have to forcibly terminate it by quitting the terminal app.

    The OpenCV version installed is 3.3.0 and the Python version is 2.7.14

    My command line execution statement is:
    python –image /images/cat_01.jpg

    I’m new to mac as well as python. There’s no error that is being shown that I can check. And I checked if I’m passing the arguments correctly. Not sure what this is about. Would really appreciate any help.

    And thanks for putting together such a great tutorial!

    • Adrian Rosebrock September 23, 2017 at 10:01 am #

      Can you confirm that both cv2.imshow and cv2.waitKey are in your code? Both of these functions are required in order to see the output on your screen.

Leave a Reply