Improving Quality of Photos on the Ubcoin Marketplace Using Neural Networks

To improve quality of photos uploaded by users, Ubcoin Market team implements Super Resolution technology. After a listing is created by a user, the image is automatically sent to the artificial intelligence module, where it is processed by the neural network to improve the image quality and make the listing more attractive to potential buyers.

The image quality of the product affects the level of sales in marketplaces. Often large manufacturers and experienced sellers have the advantage of being able to create high quality content. However, smaller sellers are often limited by the quality of their cameras on mobile phones. Ubcoin Market iis improving low image quality using Super Resolution technology.

There are four main approaches to image enhancement: prediction models, edge based methods, image statistical methods and patch based (or example-based) methods. The best quality is provided by patch based (or example-based) methods.

In [1], it was first proposed to move away from the traditional approach that applies a set of filters to a single convolutional neural network that operates on the end-to-end principle.

The approach based on the convolutional neural network combines the following functions:

1) Extracting and displaying patches.

2) Nonlinear mapping.

3) Reconstruction.

The approach based on the convolutional neural network allowed to improve the quality in comparison with the known methods while maintaining a high speed of operation.

Deep neural network architectures [2] gives the ability to restore the “spoiled image”. The solution for restoring “spoiled images” is used in the Ubcoin Market to remove watermarks and other image defects.

The application of the GAN (Generative adversarial network) allowed to improve the quality of the texture and made the processed images so photorealistic that it is visually difficult to distinguish them from the original.

Super Resolution technology is already used in image and video processing. For example, Yandex improved the quality of old movies, and some video companies use Super Resolution to improve the quality of images in cloud-based video surveillance so that customers can use more simple and cheaper cameras while preserving image quality.

The use of Super Resolution technology does not require any additional actions from Ubcoin Market user. The image from a freshly created listing is automatically sent to the artificial intelligence module, where it is processed by the neural network to improve the image quality and to make the listing more attractive to potential buyers.

[1] arxiv.org/pdf/1501.00092v3.pdf

Image Super-Resolution Using Deep Convolutional Networks

[2] arxiv.org/pdf/1606.08921.pdf

Image Restoration Using Convolutional Auto-encoders with Symmetric Skip Connections