陰影去除（Shadow Removal）是一種影像處理技術，旨在從影像中消除或減輕由光源投射的影子造成的影響。在攝影、計算機視覺和圖像處理領域，影子是一個常見的問題，因為它們可能導致圖像的對比度降低、細節的丟失以及視覺上的混淆。陰影去除的目標是恢復或修復受陰影影響的區域，以提高圖像的視覺質量和識別能力。這項任務通常需要利用圖像處理技術和計算機視覺算法來分析和處理圖像。常見的陰影去除方法分別有基於顏色轉換、基於物體分割、基於統計建模、基於物理模型、以及當今流行的基於機器學習的方法。
本篇研究將採用深度學習的方法，通過大量的陰影和非陰影圖像對，學習陰影去除的映射函數，現有的方法在處理陰影區域與非陰影區域時大多採用相同的運算操作，卻忽略了兩者之間的顏色映射本身就存在著巨大差距，直觀地來看這會使模型重建圖像的質量不佳，為了解決這個問題，本論文提出了多分支的卷積層操作，將陰影區域與非陰影區域分別計算，降低了兩者之間的依賴性，並結合了形態學運算，從而提高模型的性能與準確性。

Shadow removal is an image processing technique aimed at eliminating or reducing the impact of shadows cast by light sources in an image. Shadows are a common issue in photography, computer vision, and image processing, as they can result in reduced contrast, loss of details, and visual confusion. The objective of shadow removal is to restore or repair the areas affected by shadows to improve the visual quality and recognition capabilities of the image. This task typically involves the utilization of image processing techniques and computer vision algorithms to analyze and process the image.

Common methods for shadow removal include color-based transformations, object segmentation-based approaches, statistical modeling, physical models, and the currently popular machine learning-based methods. In this research, we will adopt a deep learning approach to learn the mapping function for shadow removal through a large dataset of shadow and non-shadow image pairs. Existing methods often apply similar computational operations for both shadow and non-shadow regions, disregarding the substantial differences in color mappings between the two. Intuitively, this can lead to poor quality when reconstructing the image. To address this issue, this paper proposes a multi-branch convolutional layer operation that separately processes shadow and non-shadow regions, reducing their interdependence. Additionally, morphological operations are incorporated to enhance the model's performance and accuracy.

[1] Alex Krizhevsky, Ilya Sutskever, Geoffrey E. Hinton, “ImageNet Classification with Deep Convolutional Neural Networks,” Part of Advances in Neural Information Processing Systems 25 (NIPS 2012), p. 9, 6 2012.
[2] N. Akhtar, A. Mian, “Threat of adversarial attacks on deep learning in computer vision: A survey,” IEEE Access, 2018.
[3] Li Xu, Jiaya Jia, Wen Gao, “Shadow Removal from a Single Image Using Global Matting,” IEEE Transactions on Pattern Analysis and Machine Intelligence, 2012.
[4] Xiao Tan, Huchuan Lu, Ming-Hsuan Yang, Shuicheng Yan, “Shadow Removal Using Local Tone Mapping and Luminance Gradient,” IEEE Transactions on Image Processing, 2012.
[5] Kaiming He, Jian Sun, Xiaoou Tang, “Robust Shadow Removal via Automatic Shadow Model Estimation,” IEEE Transactions on Image Processing, 2011.
[6] Ying-Cong Chen, Xiaowu Chen, Jiaya Jia, “Shadow Removal via Shadow-Aware Non-local Sparse Coding,” IEEE Transactions on Pattern Analysis and Machine Intelligence, 2013.
[7] Guang Chen, Yuanhao Gong, Kang Chen, “Shadow Detection and Removal using Deep Convolutional Neural Networks,” IEEE Transactions on Image Processing, 2017.
[8] Huynh Van Anh, Yeong Jun Koh, Hyun Seung Yang, “ShadowGAN: Shadow Removal with Generative Adversarial Networks,” IEEE International Conference on Computer Vision (ICCV), 2019.
[9] Yann LeCun, Bernard Boser, Joshua S. Denker, Donnie Henderson, Richard E. Howard, Wayne E. Hubbard, Lawrence D. Jackel, “Backpropagation Applied to Handwritten Zip Code Recognition,” Neural Computation, 1990.
[10] Xiaodong Cun, Chi-Man Pun, Cheng Shi, “Towards Ghost-free Shadow Removal via Dual Hierarchical Aggregation Network and Shadow Matting GAN,” Association for the Advancement of Artificial Intelligence(AAAI), 2020.
[11] Karen Simonyan, Andrew Zisserman, “Very Deep Convolutional Networks for Large-Scale Image Recognition,” IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2015.
[12] Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun, “Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition,” IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 2015.
[13] Jie Hu, Li Shen, Gang Sun, “SENet: Squeeze-and-Excitation Networks,” IEEE Transactions on Pattern Analysis and Machine Intelligence（TPAMI）, 2018.
[14] Mingxing Tan, Quoc V. Le, “EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks,” International Conference on Machine Learning（ICML）, 2019.
[15] S. K. Pal, P. Mitra, “Image Analysis by Mathematical Morphology,” IEEE Transactions on Pattern Analysis and Machine Intelligence, 1983.
[16] Kaiwen Guo, Qi Ma, Jingyi Yu, Haibin Ling, “Shadow Detection and Removal from a Single Image,” Conference on Computer Vision and Pattern Recognition(CVPR), 2019.
[17] Hieu Le, Dimitris Samaras, “Shadow Removal via Shadow Image Decomposition,” arXiv, 2019.
[18] Jifeng Wang, Xiang Li, and Jian Yang, “Stacked conditional generative adversarial networks for jointly learning shadow detection and shadow removal,” Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018.
[19] Justin Johnson, Alexandre Alahi, and Li Fei-Fei, “Perceptual Losses for Real-Time Style Transfer and Super-Resolution,” European Conference on Computer Vision (ECCV), 2016.
[20] Zhou Wang, Alan C. Bovik, Hamid R. Sheikh, Eero P. Simoncelli, “Image Quality Assessment: From Error Visibility to Structural Similarity,” IEEE Transactions on Image Processing, 7 2004.
[21] Jonathan Long, Evan Shelhamer, Trevor Darrell, “Fully Convolutional Networks for Semantic Segmentation,” Conference on Computer Vision and Pattern Recognition (CVPR), 2015.
[22] Ross Girshick, Jeff Donahue, Trevor Darrell, Jitendra Malik, “R-CNN: Rich feature hierarchies for accurate object detection and semantic segmentation,” International Conference on Computer Vision (ICCV), 2014.
[23] Guojun Yin, Chunping Hou, Shuqiang Jiang, “Single Image Shadow Detection and Removal using Paired Adversarial Networks,” IEEE Transactions on Image Processing, 2019.
[24] Zhang, Ng, and Chen, “Single Image Reflection Separation with Perceptual Losses,” CVPR, 2018.