Remember those photographs taken in black and white? Can you imagine now what they would look like in colour? Well, Artificial Intelligence algorithms, Computer Vision y Machine Learning are making it possible to colourise those memorable memories.
COLOURISATION AND NEURAL NETWORKS
The process of colourising images first requires specialists who know or imagine the context in which the black-and-white photographs were taken and who know how to use specialised design software such as Photoshop. Then, for days or even weeks, these specialists turn to a combination of expertise and patience to properly colourise each portion of the photograph.
On the other hand, neural networks, since their invention in the 1950s, have made it possible to automate a series of processes in different industries. And the process of colourisation has not been exempt from this. A variant of this algorithm called Generative Adversarial Networks (GANs) allows any given image to be coloured in black and white in a matter of seconds.
HOW DOES IT WORK?
The algorithm is composed of two neural networks. The first network aims to generate an image with colour pixels given the black and white image and the second one tries to discriminate the generated colour image versus the original colour image. Thus during several learning iterations, the generating network will be able to convince the discriminating network that the generated images are as realistic as the original ones. When this happens, the GAN will have learned to colourise various shades of grey from the black and white image.
More examples ...
A few weeks ago this tool was used to colourise black and white photographs taken by Martin Chambi. The result is incredible, as the colourisation of the images resembles the work done by an expert using specialised software for several days.
Neural Networks are the foundation of Deep Learning, the cornerstone of Machine Learning and Artificial Intelligence. Their use is spreading to many industries and is promising to automate various processes, such as colourisation, reliably and accurately.
Let's talk. Contact us for further information.