基於深度學習之具多樣性色彩映射演算法

Abstract

色彩映射在高動態範圍成像(High Dynamic Range Imaging)中扮演著一個重要角色，其用於在保留視覺特徵與美觀的特性下壓縮一影像由高動態範圍(High Dynamic Range,HDR)至低動態範圍(Low Dynamic Range,LDR)以便於呈現在顯示器之上。過去雖然有許多優秀的色彩映射演算法，但其往往只能呈現一個特定預先設計的風格且於不同場景合適的演算法會有所不同，而且，演算法的優缺評價是很主觀的，也會隨著不同的人而變化，因此，本論文提出一使用深度學習(Deep Learning)且基於傳統架構的色彩映射(Tone Mapping)演算法，並使用BicycleGAN的訓練架構，令此演算法具有生成不同風格的特性，使用者僅需更改隱性分類碼(Latent code)，即可簡易的取得個人喜好的結果。建立在傳統方法的生成器(Generator)架構幫助我們減少生成對抗網絡(Generative Adversarial Network,GAN訓練中的不確定性，產生美觀、無瑕疵(artifact)的輸出。最後，我們對此方法進行檢測，並且在主觀與客觀的品質指標得到十分優秀的成績，皆優於目前存在的傳統或深度學習色彩映射演算法。

Tone-mapping plays an essential role in high dynamic range (HDR) imaging.It aims to preserve visual information of HDR images in a medium with a limited dynamic range.Although many works have been proposed to provide tone-mapped results from HDR images, most of them can only perform tone-mapping in a single pre-designed way.However, the subjectivity of tone-mapping quality varies from person to person, and the preference of tone-mapping style also differs from application to application.In this paper, a learning-based multimodal tone-mapping method is proposed, which not only achieves excellent visual quality but also enables easy style adjustability. Based on the framework of an improved cVAE-GAN, the proposed method can provide a variety of expert-level tone-mapping results by manipulating different latent codes. Moreover, the proposed method is fast and of minimal artifacts among both learning based and non-learning based methods.The tone-mapped visual quality also outperforms the stat-of-the-art tone-mapping algorithms quantitatively and qualitatively.