Cloudinary Blog

Automatic face-blurring in images made easy

Automatically Blur Faces by Using the Pixelate Effect on Photos
After our recent, somewhat technical posts, we wanted to lighten things up with Cloudinary's latest (cool) semantic image transformation feature. Keeping people privacy in photos by automatically blurring their faces. 
 

Pixelate Effect

We previously detailed in length about Cloudinary's cloud-based API for applying effects on images. We now want to introduce Cloudinary’s latest effect - pixelization.
 
 
Use it by simply setting the 'effect' parameter (or 'e' for URLs) to 'pixelate'.
 
The example below generated a 150x200 thumbnail of a Facebook profile picture and applied the pixelization effect on it.
 
.../image/facebook/w_150,h_200,c_fill,e_pixelate/zuck.jpg
 
 
 
Isn't that great? With zero effort, you've converted an image to a pixelated one before embedding it in your site.
 
page load speed
 
As you can see, the pixelate effect in the example above is quite intense. You can assume greater control on the effect’s intensity by specifying the size of the pixelization squares (numeric value, in pixels). In the following example, we applied the pixelate effect using six pixel blocks on the same Facebook profile picture that was automatically retrieved from Facebook by Cloudinary. As always, the resulting image is cached and delivered through a fast CDN.
 
.../image/facebook/w_150,h_200,c_fill,e_pixelate:6/zuck.jpg
 
 
 
Same example in Ruby on Rails, Django, Node.js and PHP:
 
Copy to clipboard
<%= facebook_profile_image_tag("zuck.jpg", :effect => "pixelate:6", 
                               :width => 150, :height => 200, :crop => :fill) %gt;
Copy to clipboard
{% cloudinary "zuck.jpg" type="facebook" effect="pixelate:6" 
              width=150 height=200 crop="fill"  %}
Copy to clipboard
cloudinary.image("zuck.jpg", { type: "facebook",  effect: "pixelate:6", 
                 width: 150, height: 200, crop: "fill"})
Copy to clipboard
<?php echo facebook_profile_image_tag("zuck.jpg", array("effect" => "pixelate:6", 
                             "width" => 150, "height" => 200, "crop" => "fill")) ?>

 

Automatic Face Blurring

In the examples above we've shown how to easily pixelate a whole picture. But if we are talking about privacy, showing the original image as is and just hiding the faces is probably more useful.
 
For example, let's take the following photo, depicting a nice couple:
 
 
 
 
We can now apply the new 'pixelate_faces' effect to hide the faces of the couple in the original image. The following URL generates a 200x200 rounded cornered thumbnail that was smartly cropped based of face detection, while blurring the faces in the process.
 
 
 
 
As you can see, all parts of the images are still clear while the faces in the picture were automatically detected and pixelated by Cloudinary.
 
Same example in Ruby on Rails, Django, Node.js and PHP:
 
Copy to clipboard
<%= cl_image_tag("couple.jpg", :effect => "pixelate_faces", :width => 200, 
                 :height => 200, :crop => :thumb, :gravity => :faces, :radius => 20) %>
Copy to clipboard
{% cloudinary "couple.jpg" effect="pixelate_faces" width=200 height=200 
              crop="thumb" gravity="faces" radius=20  %}
Copy to clipboard
cloudinary.image("couple.jpg", { effect: "pixelate_faces", width: 200, 
                 height: 200, crop: "thumb", gravity: "faces", radius: 20})
Copy to clipboard
<?php echo cl_image_tag("couple.jpg", array("effect" => "pixelate_faces", 
                        "width" => 200, "height" => 200, "crop" => "thumb", 
                        "gravity" => "faces", "radius" => 20)) ?>
 
With the pixelate faces effect, you can pixelate one or more faces automatically. You can even pixelate the faces of a whole soccer team as seen in the example below. Here, Cloudinary automatically fetches the remote image from Wikimedia Commons and applies the pixelate faces effect before delivery through a CDN:
 
 
 
If you are a Spanish or a European football fan, you probably recognized all players in the picture despite the heavy pixelization :) 
 
As with the pixelate effect, the pixelate_faces effect also supports specifying the number of pixels to use for each pixelization block. The following example applies a 4-pixels effect on the same remote image for making it easier for you to recognize the players...
 
 
  

Face Blurring as Incoming Transformation 

For simplicity, all the examples given above were applied on the images on-demand. A user, looking at the image source, may craft a URL that does not include the blur. 
 
To prevent access to the original image, you should apply the pixelate effect as part of an incoming transforming while uploading the image to Cloudinary and before it is stored in the cloud for delivery. Alternatively, you can also upload images as private images as we described in a previous blog post

Summary 

We've had great fun introducing this feature, and personally think it's quite cool :)
 
Sites with user generated content, news sites and dating sites might find this feature very useful. Note however that automatic face detection is not perfect, and faces might not be detected and pixelated if the face in the picture is not clear enough. If you are going to incorporate face blurring in your system, you might want to introduce it as a part of a semi-automatic moderation system. By suggesting the automatically censored photo to a human moderator, you will save lengthy manual image processing time, while allowing the moderator to override incorrectly censored photos.
 
This feature was actually not in our original plans or to-do list. It was requested by one of our clients. Thankfully, the Cloudinary system was built with enough flexibility to introduce this new feature in no time. We were especially lucky to be able to reuse many existing building blocks - face detection, image effects support, best of breed image manipulation software and more.
 
If you need a yet unsupported cloud-based image transformation for your own web application, why don't you go ahead and challenge us?

Recent Blog Posts

Identifying Errors in Customer Media Assets With an Internal Tool

I still remember well my first week as a DevOps at Cloudinary. The year was 2017. Everything was new to me—people, laptop, processes—all of which to become familiar with in short order. A mantra often repeated to me in those days was that, Cloudinary being a SaaS, continuous service uptime is its most important goal.

Read more
Transformation Management Made Easy With Cloudinary

Transforming images with Cloudinary is simple, right? You find the ones you want and add them to your URL, building them up and trying them out until they’re perfect. You can then save the transformations that you created as named transformations for reuse. That final step sounds straightforward, but up until now has been a little tricky. With Cloudinary’s slick and new UI, creating and managing transformations through the Management Console just got even simpler. That UI now includes—

Read more
Generate Personalized Video Slideshows Automatically

Slideshows. Though sometimes boring, they are an essential cog in the corporate wheel, having served for decades as the medium for presentations. For a more lively display, consider combining various media types into a single video slideshow, which could serve as compelling, personalized, and engaging content for your users. Additionally, you can apply sophisticated transition configurations and Cloudinary’s many transformation capabilities.

Read more
It’s High Time to Replace JPEG With a Next-Generation Image Codec

I can be quite passionate about image codecs. A “codec battle” is brewing, and I’m not the only one to have opinions about that. Obviously, as the chair of the JPEG XL ad hoc group in the JPEG Committee, I’m firmly in the camp of the codec I’ve been working on for years. Here in this post, however, I’ll strive to be fair and neutral.

Read more
Adopting New Image Format AVIF With Cloudinary

AVIF is a new image format for the web. Before I tell you all about it, let me show you what AVIF can do.

One way to compare image codecs is to encode the same image in different formats at matched file sizes and then compare the visual quality of the resulting images. For example, I rendered the AVIF below with a q_50 quality transformation. It weighs 12.3 KB and, compared to the lossless original it looks pretty good subjectively.

Read more