AI Image / Video
With the advent of neural networks and in particular image classifier networks, the ability to manipulate images quickly and in large numbers has become a reality. As part of Trender’s service suite we now offer AI based image and video manipulation for discrete image/video projects or as part of our general web design service.
Intelligent Upscaling
The process of image up-scaling involves taking a low resolution image and feeding it into the neural network where it is analyzed and processed before being up-scaled to a higher resolution with additional pixel data. The neural network makes a best guess attempt at “filling in the blanks” to recreate the image as if it was taken on equipment capable of higher resolution capture.
In situations where an important image needs to be used but which only exists in small resolutions, AI up-scaling can offer a solution to generating usable images from poor source data.
Intelligent De-noising
The process of de-noising involves removing unwanted data from the source image, (noisy data results in a grainy effect) and filling in these areas with smoother texture and color. This technique can help restore photos that were taken on poor quality equipment or in low light conditions.
Motion Video
Much like still images, low resolution video can be up-scaled via the neural network to higher resolutions and wide screen aspect ratios. In addition to resolution up-scaling, the network can de-noise and de-block in parallel resulting in a cleaner video.
Colorisation
Intelligent image classifier technology can detect and classify objects within the frame and automatically colorise them with impressive accuracy.
Pricing
Static Image Upscale / De-noise / Colorize – $0.30 per image.
Video – Upscale / De-noise / De-block – $30.00/hr of footage.
Technical Overview and Limitations
Neural Image classifier networks are a very emergent technology and although incredible in what they’re able to achieve, still have some limitations.
For instance, where a low resolution image is up-scaled, the AI attempts to do its best at filling in areas of missing data by comparing what it’s seeing to what should be there. It then makes a best guess estimate at filling in those blanks with new data it generates itself. Sometimes this can result in a shimmery, ‘water-color painting’ effect that although higher resolution does look a little computer generated compared to the original. Most of these trade offs however are offset by the increased resolution and general tidying up of the image.
Removing noise from an image is generally quite effective however sometimes the AI can erroneously smooth areas that it should not touch and also leave things that it should have removed. It is however true to say that 95% of the time the net sum is a marked improvement over the original.
It is important to be aware of these quirks before engaging Trender to AI process your images and video footage. We will always endeavor to generate the best output quality but there are some inherent technological limitations that at the time of this writing cannot be fixed.
Contact
For additional information regarding Trender’s AI based image processing services, please click here and fill in our contact form with your query.