Search

Infrared and visible image fusion, which focuses on generating a new synthetic image, seeks to integrate the complementary properties of both modalities to contain abundant information. Many existing image fusion methods adopt artificially designed fusion rules. However, they often lack sufficient consideration for the interaction of long-range context information. As a result, important information may be lost, which limits the application of subsequent vision tasks. To address these limitations, we propose a novel unsupervised image fusion network named DCSFuse, which is guided by the correlation of image features. Our method can adaptively integrate complementary and long-range context information. More specifically, the proposed method first learns the modal-specific features of the two modalities. Then it calculates the correlation between each feature and integrates the cross-modal features guided by the correlation. Finally, these integrated features are reconstructed into a fused image that contains supplementary information from both modalities and captures long-range context dependencies. Extensive experiments on three mainstream datasets demonstrate the superiority of our method over other methods in terms of quantitative metrics and qualitative effects. Furthermore, we demonstrate experimentally that our fused images enhance the accuracy of visual object detection tasks. In particular, our network benefits from its superior computational efficiency, enabling it to generate fused images in real time. The source code of DCSFuse has been released at: https://github.com/zc617/DCSFuse.

Feature matching plays a key role in computer vision. However, due to the limitations of the descriptors, the putative matches are inevitably contaminated by massive outliers. This paper attempts to tackle the outlier filtering problem from two aspects. First, a robust and efficient graph interaction model, is proposed, with the assumption that matches are correlated with each other rather than independently distributed. To this end, we construct a graph based on the local relationships of matches and formulate the outlier filtering task as a binary labeling energy minimization problem, where the pairwise term encodes the interaction between matches. We further show that this formulation can be solved globally by graph cut algorithm. Our new formulation always improves the performance of previous locality-based method without noticeable deterioration in processing time, adding a few milliseconds. Second, to construct a better graph structure, a robust and geometrically meaningful topology-aware relationship is developed to capture the topology relationship between matches. The two components in sum lead to topology interaction matching (TIM), an effective and efficient method for outlier filtering. Extensive experiments on several large and diverse datasets for multiple vision tasks including general feature matching, as well as relative pose estimation, homography and fundamental matrix estimation, loop-closure detection, and multi-modal image matching, demonstrate that our TIM is more competitive than current state-of-the-art methods, in terms of generality, efficiency, and effectiveness. The source code is publicly available at http://github.com/YifanLu2000/TIM.

Spectral unmixing (SU), which refers to extracting basic features (i.e., endmembers) at the subpixel level and calculating the corresponding proportion (i.e., abundances), has become a major preprocessing technique for the hyperspectral image analysis. Since the unmixing procedure can be explained as finding a set of low-dimensional representations that reconstruct the data with their corresponding bases, autoencoders (AEs) have been effectively designed to address unsupervised SU problems. However, their ability to exploit the prior properties remains limited, and noise and initialization conditions will greatly affect the performance of unmixing. In this article, we propose a novel technique network for unsupervised unmixing which is based on the adversarial AE, termed as adversarial autoencoder network (AAENet), to address the above problems. First, the image to be unmixed is assumed to be partitioned into homogeneous regions. Then, considering the spatial correlation between local pixels, the pixels in the same region are assumed to share the same statistical properties (means and covariances) and abundance can be modeled to follow an appropriate prior distribution. Then the adversarial training procedure is adapted to transfer the spatial information into the network. By matching the aggregated posterior of the abundance with a certain prior distribution to correct the weight of unmixing, the proposed AAENet exhibits a more accurate and interpretable unmixing performance. Compared with the traditional AE method, our approach can greatly enhance the performance and robustness of the model by using the adversarial procedure and adding the abundance prior to the framework. The experiments on both the simulated and real hyperspectral data demonstrate that the proposed algorithm can outperform the other state-of-the-art methods.