This is a guest post by Gilad David Maayan.

Optimizing images is crucial for maintaining a good user experience on your website. Load time statistics show that if the average page load exceeds 3 seconds, the abandonment rates soar to a scary 40%. The most important factor affecting loading time is the size of the visual media files used in the website, including images, graphics and videos. According to HTTP Archive, around 64% of a website’s average weight is images. Video accounts for another 9%.

However, machine learning is changing how we create, edit and store images. Today, there are several image optimization and management solutions, including Digital Asset Management software, which use machine learning to improve image processing and optimization. Read on to learn more about the role of machine learning in image optimization tasks. 

What Is Image Optimization?

Image optimization refers to the delivery of images in the format, dimensions, resolution and quality that has the smallest possible file size while keeping the highest possible quality. One of the most powerful features of image optimization is the ability to adjust the size and quality of your images to a level that still provides good visual output. You can deliver images at less than half the original size with almost no noticeable difference. 

Optimizing your images can significantly reduce page load times, which in turn encourages visitor permanence, or the likelihood a user will stay around on your website. Moreover, Google is prioritizing faster sites in search results, so a fast loading speed results in a higher ranking. 

Image Optimization Challenges

You decided to start optimizing your images to improve SEO and loading time. However, you find yourself struggling with low-quality results or images that simply “refuse” to resize well. Optimizing images is not without its challenges. The following are some of the most common ones, and how to solve them, including: 

Low image quality

Usually, photographs are delivered at their original extra-high resolution, to keep the visual quality or to help users with high-res devices get the expected quality. This often results in a huge file, slowing the page loading time. 

Solution: You can lower the quality of the file without the image changing significantly. Use the lowest possible quality acceptable for the content, audience and purpose. 

Image format

All image formats use image compression. Some image formats are lossy, discarding some pixel data, others are lossless, which don’t discard so much data. For example, JPG is lossy, while PNG is lossless. Small images, such as buttons and logos, are often overlooked but they make up a large proportion of site images. Moreover, these types of images are usually in PNG, resulting in a large cumulative size, slowing your page load time.  

Solution: ensure each image is delivered in the right format for the content. Use newer formats such as WebP or JPEG-XR, to deliver animations instead of GIFs. Vector graphics are often more suitable for logos than PNGs, being minimal in size and delivering sharp results at every resolution and zoom. 

Image metadata

Images contain a lot of metadata, stored by cameras and graphics applications, which is unnecessary when the image is delivered.  For example, descriptive, structural and administrative information may be useful for identifying files and managing assets internally, but they are of little use to the end-user.

Solution: The best option is to remove this metadata from delivered images, but keep it in the original copy of the graphics.


If you rely on the browser to resize the images, it may deliver unnecessarily large files. This problem applies both to standard resizing and responsive design. Thus, it is important to crop your images to focus on important content. 

Solution: You can crop to focus on important content on the server-side. Using a Digital Asset Management solution can help you automate some tasks on image optimization, among other benefits. You can learn more about DAM solutions here

Image Optimization with Machine Learning

It makes perfect sense to apply machine learning to image optimization. After all, there have been a lot of improvements in both the machine learning and deep learning fields, with computers learning to recognize, optimize, compress and change image formats. The following are some of the ways machine learning is improving image optimization.

Image compression and resolution

RAISR, an algorithm released by Google, combines traditional upsampling with deep learning to change low-resolution images into high-resolution counterparts. Another company, WaveOne, has trained a model to compress images to small sizes, with more success than JPEG. 

Image Enhancement

Google did it again, with a new neural network architecture that reproduces sophisticated image enhancements with inference running in real time at full HD resolution on mobile devices. The software can also learn subjective effects from human retouching, which means that machine learning can perform human-like photo retouching in real-time on your phone. 

So, instead of relying on Instagram filters, this model shows you the final result of your photos with professional-level enhancements as you frame up your shot. 

Image Manipulation and Generation

One way to do it is to use adversarial networks, where one network teaches the other how to generate some data. This can give some exciting results, such as turning night into day or even removing rain from pictures. The ability of networks to learn from each other and generate new data is powerful, and will most likely be used for a myriad of applications, from medical imaging to special effects.

Additionally, this type of network can be used to increase resolution. For example, a Super-Resolution Generative Adversarial Network (SRGAN) can be trained to predict and fill in the missing data in low-resolution images, producing a realistic high-resolution output.  


Optimizing our website images is a necessary task to keep up with this competitive Internet environment. Machine learning algorithms and tools help automate otherwise repetitive tasks, making part of the best image optimization and Digital Asset Management software. New and innovative technologies are enhancing and changing the way we create and edit images in mind-blowing ways. The future of image optimization will only be enriched by deep learning and machine learning techniques.

Gilad David Maayan is a technology writer who has worked with over 150 technology companies including SAP, Imperva, Samsung NEXT, NetApp and Ixia, producing technical and thought leadership content that elucidates technical solutions for developers and IT leadership. Today he heads Agile SEO, the leading marketing agency in the technology industry. LinkedIn: