Deep HDR Hallucination for Inverse Tone Mapping

Inverse Tone Mapping (ITM) methods attempt to reconstruct High Dynamic Range (HDR) information from Low (LDR) image content. The dynamic range of well-exposed areas must be expanded and any missing due over/under-exposure recovered (hallucinated). majority focus on the former are relatively successful, while most attempts latter not sufficient quality, even ones based Convolutional Neural Networks (CNNs). A major factor for reduced inpainting quality in some works is choice loss function. Work Generative Adversarial (GANs) shows promising results synthesis LDR inpainting, suggesting that GAN losses can improve inverse tone mapping results. This work presents a GAN-based method hallucinates badly exposed images compares its efficacy with alternative variations. proposed quantitatively competitive state-of-the-art methods, providing good expansion plausible hallucinations saturated under-exposed areas. density-based normalisation method, targeted HDR content, also proposed, as well an data augmentation hallucination.