Cleaning up and augmenting old pictures using AI? Check!

by Feb 9, 2021News

Because nothing is more important than family, Mikael wanted to give them a Christmas present at all costs despite his limited budget. In his search for the ideal gift, he suddenly had a very budget-friendly, but heart-warming idea. Using AI technologies, he hoped to restore and enlarge an old photograph of his grandfather. Curious about the tools he used and how our AI bootcamp prepared him for this project?

Be sure to read the interview below!

Where did the idea for this project come from?
This project happened a bit by chance. It was Christmas time, a very important holiday in my family, but I had lost my job 6 months earlier and things were getting a bit difficult to handle with the unemployment funds. I wanted to be able to make small gifts in spite of the situation. My grandmother, having passed a few months before, we had to tidy up her house and we came across some very old photos of my grandparents. It was a big surprise for me because I had always known my grandfather as an old person. I was very happy to discover him as a young man because I had always known him as a stocky elderly. In these photos he was young, slender; he looked like a strong, hard-working man.”

“My mother has three sisters, and I thought it would make them happy to receive a high-quality photograph of their father when he was much younger. At first, I didn’t even think about artificial intelligence. I thought I would learn how to use photoshop and rework the image.  But as I was watching tutorials, I stumbled upon an AI video that showed me how to do what I wanted to do: upscaling the portrait without losing image quality. So this was the perfect opportunity to use what I am learning in my AI training.

Can you describe the different steps in the process?
“The course of the project was always pretty clear to me:

First step: first of all, I inquired about the different ways of importing this small image on my PC with the least possible loss of quality. I went to the printer in my street, which had a 600 DPI scanner, and made a simple black and white scan. 

Second step: once the document was imported, I wanted to clean up all imperfections in the image, which I started to do with Photoshop. But I quickly switched to an artificial intelligence technology that I found online, bringing old photos back to life. I was able to get a very satisfying result, with lots of small stains and staple marks erased by the AI.

Third step: once the document was cleaned, I turned to Photoshop to make enlargements, apply filters, etc. I wanted to make this photo go from ID format to A4. With the help of tutorials, I managed to get something of more or less good quality, but it wasn’t incredible. And then I discovered this video that explained how to use AI for my project. I installed everything that needed to be installed, and unfortunately for me, I discovered that the image process was too much for my computer to take. So I had to find other solutions that had not yet been covered in the training. I used Google Colab, which allows you to use the CPU of their servers to do calculations. This allowed me to upscale my image. 

Fourth step: I used filters to go even further.  For example, I wanted to try and color the image. Once you get into the field of imaging, the possibilities are endless. I found AIs that allow you to add color, but it’s a technology that is only being explored and in my case, the result was not very convincing. But I have also found AIs that embellish images. I got my final result using 

Remini (Play Store / App Store) which is an app for smartphones. I discovered through my research that many photographers were pleasantly surprised by the results obtained with this app. So I tried my luck with this project and the result was incredible: it made the photo even more vivid.

Fifth step: once I was satisfied with the result, all I had to do was to print it! I plan to repeat the experience with a photo of my grandmother.”

Could you tell us a bit about the tools that you used as part of this project?
“First of all, there was Photoshop, which allowed me to tackle the subject but was not enough. Then there was Bringing old photos back to life. This tool was developed by Microsoft and is open source and free to use. Thanks to it, I was able to remove some stains and yellowing. The other big tool I used was actually the most important in the process. It’s called DFDNet and it’s an open-source project that focuses only on portraits and not on simple images. For example, this tool doesn’t deal with the background of the image. It’s a very interesting tool that uses a lot of things we saw in the training. First, it uses a convolutional neural network that takes the image, applies thousands of filters to it to see what each part of the image contains and it will determine where the eyes, nose, mouth, etc. are located. Based on this, it will search its database for the closest match. Once he has found the information, it will be able to upscale the picture part by part. It’s the best tool I’ve found to do this task in A4.  When you zoom in, you can see that a lot of pixels have been polished. I am really very happy with the result.”

Which skills learned at BeCode have been the most useful to you as part of the project?
“BeCode teaches us how to use GitHub and how to write good documentation. So it’s very nice to see other projects where the documentation is excellent. Of course, Python was very useful, as was Google Colab, even if we only saw it later in the training.

”Learning how to do good online searches, which is inherent to the active pedagogy of BeCode, was a very important factor. It teaches us to find solutions when we are stuck and to be more creative. When I was stuck because of the hardware, I searched and found Google Colab, and it worked. If it hadn’t worked, I would have had to turn to someone with more advanced hardware than me.”

Are you planning to work on similar projects in the future?
“Not at the moment. Before this experience, I had never really had a project of my own. I’m a jack-of-all-trades and always have lots of little ideas, but never a fixed goal. Here, having one has helped me a lot. I’m open to finding another goal, but it was all very intense. I worked on it every night, often late into the night. But it was very motivating because I could see the progress I was making. In fact, when it stopped evolving, I would change direction to move forward. That’s how I went from Photoshop to AI.

All in all, this project took me 3-4 weeks. I would be happy to do this kind of project again, but after my training and internship.”

Are you interested in following the AI Bootcamp?

That’s possible! – New classes are starting in Antwerp, Ghent, Brussels, Liège and Charleroi.

 

Latest news

Recent articles

Web dev coach Charleroi

Web dev coach Charleroi

Job Overview At BeCode, we have 1 mission: enabling tomorrow’s digital talents to blossom. We believe that education makes anything possible. Since 2017, BeCode has been offering inclusive coding bootcamps to jobseekers to start their career in the ICT sector as a web...