Deep Tone-Mapping Operator Using Image Quality Assessment Inspired Semi-Supervised Learning

Tone-mapping operator (TMO) is intended to convert high dynamic range (HDR) content into a lower so that it can be displayed on standard (SDR) device. The tone-mapped result of HDR usually stored as SDR image. For different scenes, traditional TMOs are able obtain satisfying image only under manually fine-tuned parameters. In this paper, we address problem by proposing learning-based TMO using deep convolutional neural network (CNN). We explore CNN structure and adopt multi-scale multi-branch fully design. When training CNN, introduce quality assessments (IQA), specifically, assessment, implement semi-supervised loss terms. discuss prove the effectiveness terms, structure, data pre-processing, etc. several experiments. Finally, demonstrate our approach produce appealing results diversified scenes.