Enhance Image Quality using AI

Enhance (verb): to raise to a higher degree; intensify; magnify. (dictionary.com)

AI imaging techniques can be used to enhance image quality. In the previous section, we took a look at AI Image editing, Now you might wonder, what is the difference between AI Image Editing and AI based Image enhancement?

#AI Image editing

In order to get a clarity on the two terms, let us re visit what AI Image editing can do. This technique helps us in removing objects from an image, changing the background, or touching up the appearance of the people in the images (e.g. red eye reduction). Other modifications may include adding text, layers or filters, cropping or rotating the image, etc.

#AI Image enhancement

This refers to using AI techniques to improve the image quality or resolution of an image. This may include one or more of the following: changing the brightness or contrast, reducing the 'noise' in the image; or changing the resolution (in particular, up-scaling the image) are some of the manipulations or modifications to the images.

Indeed one can get confused trying to differentiate between the two. This may be particularly true for those who are not too comfortable with the use of the English language. For now, let us keep the definitions aside and focus on image enhancement using AI.

#De-noising an image using AI

#Sites that enhance images using AI

Many sites and apps tend to promote the image upscaling feature of AI based image enhancement. You may find the list of such sites in many blog posts, newsletters, and listicles. Here is one such post. Also note that just like AI Image generation, there are different techniques for AI image Enhancement. Below is a screenshot from the Image enhancement or upscaling tool of Supermachine.

I tested three random sites as a part of writing this section. These sites are:

• Vanceai
• img2go, and
• upscale.media

Each of these sites allow a free trial, and since my main interest was to test the concept or idea anyways, I opted for the free trial. I could upscale a test image to 2x and then 4x on two of the sites (Vance and upscale.media), while in img2go I could pre-define the resolution (width x height, in pixels).

For my test with these three sites, I chose an image with a very small resolution, about 315 x 315 pixels which was generated using art.elbo. While most image enhancement sites accept .tiff, .jpg and .png formats; I opted for downloading the images as .jpg because of my personal preference.

I uploaded the image to each of the three sites, and chose 2x or 4x increase in size. On img2go, I could define the size in pixels, so I opted for 1000x1000 pixels, or more than a 3x increase in image resolution.

The results were rather disappointing- I could see pixels below the eyes of the woman in the image for one, and in general, the image did not look very sharp. It might have to do with the low resolution of the base image. Or maybe something else.

#Summarizing the results

Based on a very quick test and a limited data set, below are the results. Bear in mind that my objective is to illustrate how AI image enhancement is and what it can do, rather than discussing the merits of AI image enhancement technology.

I was pleasantly surprised to find that the img2go delivered image with 3x the resolution, but the file size was 80 percent of the original! Time permitting, I might run a more comprehensive test. For now, I believe the above table gives you a fair idea about the impact on size of image vis-a-vis resolution, and keep in mind that image quality trumps size of the image!

Now that we have reached the end of this section, I think it is important to re-iterate the differences between Image Editing and Image Enhancement:

For the sake of simplicity, let us consider the following: Image Enhancement means improving the image quality. This may include increasing the resolution, brightness, contrast, and colour balance. Most other modifications to an image fall under the category of image editing.