25 May 2020 in IEEE Transactions on Pattern Analysis and Machine Intelligence
Multi-view stereopsis (MVS) tries to recover the 3D model from 2D images. As the observations become sparser, the significant 3D information loss makes the MVS problem more challenging. Instead of only focusing on densely sampled conditions, we investigate sparse-MVS with large baseline angles since the sparser sensation is more practical and more cost-efficient. By investigating various observation sparsities, we show that the classical depth-fusion pipeline becomes powerless for the case with a larger baseline angle that worsens the photo-consistency check. As another line of the solution, we present SurfaceNet+, a volumetric method to handle the ‘incompleteness’ and the ‘inaccuracy’ problems induced by a very sparse MVS setup. Specifically, the former problem is handled by a novel volume-wise view selection approach. It owns superiority in selecting valid views while discarding invalid occluded views by considering the geometric prior. Furthermore, the latter problem is handled via a multi-scale strategy that consequently refines the recovered geometry around the region with the repeating pattern. The experiments demonstrate the tremendous performance gap between SurfaceNet+ and state-of-the-art methods in terms of precision and recall. Under the extreme sparse-MVS settings in two datasets, where existing methods can only return very few points, SurfaceNet+ still works as well as in the dense MVS setting.
1 June 2020 in Nature Machine Intelligence
18 September 2024 in CAAI Transactions on Intelligence Technology
Images with complementary spectral information can be recorded using image sensors that can identify visible and near‐infrared spectrum. The fusion of visible and near‐infrared (NIR) aims to enhance the quality of images acquired by video monitoring systems for the ease of user observation and data processing. Unfortunately, current fusion algorithms produce artefacts and colour distortion since they cannot make use of spectrum properties and are lacking in information complementarity. Therefore, an information complementarity fusion (ICF) model is designed based on physical signals. In order to separate high‐frequency noise from important information in distinct frequency layers, the authors first extracted texture‐scale and edge‐scale layers using a two‐scale filter. Second, the difference map between visible and near‐infrared was filtered using the extended‐DoG filter to produce the initial visible‐NIR complementary weight map. Then, to generate a guide map, the near‐infrared image with night adjustment was processed as well. The final complementarity weight map was subsequently derived via an arctanI function mapping using the guide map and the initial weight maps. Finally, fusion images were generated with the complementarity weight maps. The experimental results demonstrate that the proposed approach outperforms the state‐of‐the‐art in both avoiding artificial colours as well as effectively utilising information complementarity.
23 August 2023 in Engineering
23 February 2021 in IEEE Transactions on Image Processing
Learning-based single image super-resolution (SISR) aims to learn a versatile mapping from low resolution (LR) image to its high resolution (HR) version. The critical challenge is to bias the network training towards continuous and sharp edges. For the first time in this work, we propose an implicit boundary prior learnt from multi-view observations to significantly mitigate the challenge in SISR we outline. Specifically, the multi-image prior that encodes both disparity information and boundary structure of the scene supervise a SISR network for edge-preserving. For simplicity, in the training procedure of our framework, light field (LF) serves as an effective multi-image prior, and a hybrid loss function jointly considers the content, structure, variance as well as disparity information from 4D LF data. Consequently, for inference, such a general training scheme boosts the performance of various SISR networks, especially for the regions along edges. Extensive experiments on representative backbone SISR architectures constantly show the effectiveness of the proposed method, leading to around 0.6 dB gain without modifying the network architecture.