Discriminative feature encoding for intrinsic image decomposition
Abstract Intrinsic image decomposition is an important and long-standing computer vision problem. Given an input image, recovering the physical scene properties is ill-posed. Several physically motivated priors have been used to restrict the solution space of the optimization problem for intrinsic i...
Main Authors: | , , |
---|---|
Format: | Article |
Language: | English |
Published: |
SpringerOpen
2023-04-01
|
Series: | Computational Visual Media |
Subjects: | |
Online Access: | https://doi.org/10.1007/s41095-022-0294-4 |
_version_ | 1797822711546773504 |
---|---|
author | Zongji Wang Yunfei Liu Feng Lu |
author_facet | Zongji Wang Yunfei Liu Feng Lu |
author_sort | Zongji Wang |
collection | DOAJ |
description | Abstract Intrinsic image decomposition is an important and long-standing computer vision problem. Given an input image, recovering the physical scene properties is ill-posed. Several physically motivated priors have been used to restrict the solution space of the optimization problem for intrinsic image decomposition. This work takes advantage of deep learning, and shows that it can solve this challenging computer vision problem with high efficiency. The focus lies in the feature encoding phase to extract discriminative features for different intrinsic layers from an input image. To achieve this goal, we explore the distinctive characteristics of different intrinsic components in the high-dimensional feature embedding space. We define feature distribution divergence to efficiently separate the feature vectors of different intrinsic components. The feature distributions are also constrained to fit the real ones through a feature distribution consistency. In addition, a data refinement approach is provided to remove data inconsistency from the Sintel dataset, making it more suitable for intrinsic image decomposition. Our method is also extended to intrinsic video decomposition based on pixel-wise correspondences between adjacent frames. Experimental results indicate that our proposed network structure can outperform the existing state-of-the-art. |
first_indexed | 2024-03-13T10:13:10Z |
format | Article |
id | doaj.art-01da38f739cb4f7bb5b4e407762fbfd9 |
institution | Directory Open Access Journal |
issn | 2096-0433 2096-0662 |
language | English |
last_indexed | 2024-03-13T10:13:10Z |
publishDate | 2023-04-01 |
publisher | SpringerOpen |
record_format | Article |
series | Computational Visual Media |
spelling | doaj.art-01da38f739cb4f7bb5b4e407762fbfd92023-05-21T11:23:06ZengSpringerOpenComputational Visual Media2096-04332096-06622023-04-019359761810.1007/s41095-022-0294-4Discriminative feature encoding for intrinsic image decompositionZongji Wang0Yunfei Liu1Feng Lu2Key Laboratory of Network Information System Technology (NIST), Aerospace Information Research Institute, Chinese Academy of SciencesState Key Laboratory of Virtual Reality Technology and Systems, School of Computer Science and Engineering, Beihang UniversityState Key Laboratory of Virtual Reality Technology and Systems, School of Computer Science and Engineering, Beihang UniversityAbstract Intrinsic image decomposition is an important and long-standing computer vision problem. Given an input image, recovering the physical scene properties is ill-posed. Several physically motivated priors have been used to restrict the solution space of the optimization problem for intrinsic image decomposition. This work takes advantage of deep learning, and shows that it can solve this challenging computer vision problem with high efficiency. The focus lies in the feature encoding phase to extract discriminative features for different intrinsic layers from an input image. To achieve this goal, we explore the distinctive characteristics of different intrinsic components in the high-dimensional feature embedding space. We define feature distribution divergence to efficiently separate the feature vectors of different intrinsic components. The feature distributions are also constrained to fit the real ones through a feature distribution consistency. In addition, a data refinement approach is provided to remove data inconsistency from the Sintel dataset, making it more suitable for intrinsic image decomposition. Our method is also extended to intrinsic video decomposition based on pixel-wise correspondences between adjacent frames. Experimental results indicate that our proposed network structure can outperform the existing state-of-the-art.https://doi.org/10.1007/s41095-022-0294-4intrinsic image decompositiondeep learningfeature distributiondata refinement |
spellingShingle | Zongji Wang Yunfei Liu Feng Lu Discriminative feature encoding for intrinsic image decomposition Computational Visual Media intrinsic image decomposition deep learning feature distribution data refinement |
title | Discriminative feature encoding for intrinsic image decomposition |
title_full | Discriminative feature encoding for intrinsic image decomposition |
title_fullStr | Discriminative feature encoding for intrinsic image decomposition |
title_full_unstemmed | Discriminative feature encoding for intrinsic image decomposition |
title_short | Discriminative feature encoding for intrinsic image decomposition |
title_sort | discriminative feature encoding for intrinsic image decomposition |
topic | intrinsic image decomposition deep learning feature distribution data refinement |
url | https://doi.org/10.1007/s41095-022-0294-4 |
work_keys_str_mv | AT zongjiwang discriminativefeatureencodingforintrinsicimagedecomposition AT yunfeiliu discriminativefeatureencodingforintrinsicimagedecomposition AT fenglu discriminativefeatureencodingforintrinsicimagedecomposition |