Learned Perceptual Image Enhancement

TL;DR

This paper introduces a learned perceptual loss based on a CNN trained on aesthetic preferences, which improves the quality of image enhancement operators like tone mapping and dehazing by aligning results more closely with human perception.

Contribution

The paper proposes a novel perceptual loss function using a CNN trained on aesthetic data, enhancing the training of image enhancement operators without increasing inference complexity.

Findings

Perceptual loss improves enhancement quality according to human aesthetic preferences.

The method effectively tunes various operators such as tone mapping and dehazing.

The approach does not add complexity during inference, only during training.

Abstract

Learning a typical image enhancement pipeline involves minimization of a loss function between enhanced and reference images. While L1 and L2 losses are perhaps the most widely used functions for this purpose, they do not necessarily lead to perceptually compelling results. In this paper, we show that adding a learned no-reference image quality metric to the loss can significantly improve enhancement operators. This metric is implemented using a CNN (convolutional neural network) trained on a large-scale dataset labelled with aesthetic preferences of human raters. This loss allows us to conveniently perform back-propagation in our learning framework to simultaneously optimize for similarity to a given ground truth reference and perceptual quality. This perceptual loss is only used to train parameters of image processing operators, and does not impose any extra complexity at inference time. Our experiments demonstrate that this loss can be effective for tuning a variety of operators such as local tone mapping and dehazing.