Universal Filter-Based Lightweight Image Enhancement Model with Unpaired Learning Mode

Abstract

Image enhancement is crucial in digital image processing to improve visual quality across various applications. Recent advancements in deep learning and computer vision have significantly advanced automatic color correction. While heavyweight solutions excel in quality, they demand substantial computational resources, whereas emerging lightweight models promise efficient operation on mobile devices. This study introduces a lightweight neural network model suitable for mobile devices for image color gamut correction. Our model demonstrates performance comparable with heavyweight models. We propose an approach that integrates unsupervised learning methods with multimodal visual-language priors. To our knowledge, this is the first study to use multimodal architectures as a discriminator for automatic image color correction. Also, we proposed a method for evaluating the quality of Image Enhancement models based on unpaired data using binary questions answering.

Keywords

• image enhancement
• filter-based model
• unpaired learning