On May 13, 2025, content artificial intelligence (AI) solution provider 4by4 announced that it had completed the registration of a patent for a training method for an AI model and a color correction methodology that radically simplifies the color correction workflow.
The patent is titled "Training Method for Artificial Neural Network Model for Image Color Correction and Image Color Correction Method Using the Trained Artificial Neural Network Model." Unlike conventional color correction AI models that directly correct images, the company has implemented a lightweight model that precisely predicts only the core parameters required for color correction.
While traditional color correction methods output a "raw image → corrected image" directly, the model developed by 4by4 adopts a structure of "raw image → correction parameter prediction → application to image." This approach utilizes a lighter AI model, while also enhancing stability and maximizing correction processing efficiency.
The patent's utility can be summarized in three points. First, since the color correction algorithm repeatedly uses quantified information (parameters) as is, these are excluded from the training targets. As a result, the AI model's training speed and stability are greatly improved, the training process is simplified, the output becomes more consistent, and the risks of overfitting or failure are reduced.
Second, because only a small number of correction parameters need to be predicted rather than the entire image, the amount of training data and time required is significantly reduced, and prediction precision is also increased. In addition, the predicted parameters can be finely adjusted by the operator during post-processing.
Third, since the predicted correction parameters are provided in numerical form, they are useful for precisely matching specific styles in subsequent correction tasks. This enables the model to meet different color correction standards required by various production environments such as broadcasting, film, and digital content.
Bae Sungwan, Head of the Pixel Business Division at 4by4, explained, "This patent serves as an opportunity to make the AI color correction process much lighter and faster, increasing the potential for launching customized products that allow users to select only the necessary features."
He added, "Even in specialized fields such as professional video editing tools or post-production for broadcasting and film, it is now possible to ensure consistency in correction quality with just one lightweight model, without the burden of high costs. Based on this patent, we will further enhance the marketability of our image quality enhancement AI solution, Pixel."
The patent was jointly developed and applied for by 4by4 and its affiliate AI Nation, with a share ratio of 9 to 1.
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.
