面向复杂退化场景的视频逆色调映射算法

Video Inverse Tone Mapping for Complex Degradation Scenarios

摘要: 针对现有视频逆色调映射方法对降质过程先验依赖较强且泛化能力不足的问题，文章提出一种降质类型自适应的视频逆色调映射方法，以提升复杂退化场景下的高动态范围重建性能。该方法首先构建包含多种色调映射算子的训练数据集，并利用分类器对输入视频的降质特性进行预测，通过嵌入编码将类别信息引入映射过程。在此基础上，设计全局色彩变换模块实现初步动态范围扩展，引入曝光引导的空间注意力机制对过曝与欠曝区域进行细节恢复，同时采用时空特征协同对齐策略融合多帧上下文信息。实验结果表明，在所构建的包含11类退化形式的视频数据集上，所提方法在峰值信噪比、结构相似性及色度差异等指标上均优于现有主流方法。

Abstract: Existing inverse tone mapping methods heavily rely on prior knowledge of degradation processes, which limits their generalization across diverse scenarios. To address this issue, this paper proposes a degradation-type adaptive video inverse tone mapping network to improve performance under complex degradation conditions. Specifically, the proposed method constructs training data covering multiple tone mapping operators, employs a classifier to identify degradation characteristics of input videos, and embeds the predicted category information into the mapping process via an embedding encoding mechanism. Within the mapping pipeline, a global color transformation module is designed to perform initial dynamic range expansion, while an exposure-guided spatial attention mechanism is introduced to restore over-exposed and under-exposed regions. In addition, a spatio-temporal feature collaborative alignment strategy is adopted to aggregate multi-frame contextual information. Experimental results demonstrate that, on the proposed video dataset containing 11 degradation types, the proposed method outperforms existing state-of-the-art approaches in terms of peak signal-to-noise ratio (PSNR), structural similarity (SSIM), and chrominance difference.