Infrared and visible image fusion technique is a popular topic in image analysis because it can integrate complementary information and obtain reliable and accurate description of scenes. Multiscale transform theory as a signal representation method is widely used in image fusion. In this paper, a novel infrared and visible image fusion method is proposed based on spectral graph wavelet transform (SGWT) and bilateral filter. The main novelty of this study is that SGWT is used for image fusion. On the one hand, source images are decomposed by SGWT in its transform domain. The proposed approach not only effectively preserves the details of different source images, but also excellently represents the irregular areas of the source images. On the other hand, a novel weighted average method based on bilateral filter is proposed to fuse low- and high-frequency subbands by taking advantage of spatial consistency of natural images. Experimental results demonstrate that the proposed method outperforms seven recently proposed image fusion methods in terms of both visual effect and objective evaluation metrics.
Detecting key frames in videos is a common problem in many applications such as video classification, action recognition and video summarization. These tasks can be performed more efficiently using only a handful of key frames rather than the full video. Existing key frame detection approaches are mostly designed for supervised learning and require manual labelling of key frames in a large corpus of training data to train the models. Labelling requires human annotators from different backgrounds to annotate key frames in videos which is not only expensive and time consuming but also prone to subjective errors and inconsistencies between the labelers. To overcome these problems, we propose an automatic self-supervised method for detecting key frames in a video. Our method comprises a two-stream ConvNet and a novel automatic annotation architecture able to reliably annotate key frames in a video for self-supervised learning of the ConvNet. The proposed ConvNet learns deep appearance and motion features to detect frames that are unique. The trained network is then able to detect key frames in test videos. Extensive experiments on UCF101 human action and video summarization VSUMM datasets demonstrates the effectiveness of our proposed method.
27 January 2025 in IEEE Transactions on Geoscience and Remote Sensing
In a real-world infrared imaging system, effectively learning a consistent stripe noise removal model is essential. Most existing destriping methods cannot precisely reconstruct images due to cross-level semantic gaps and insufficient characterization of the global column features. To tackle this problem, we propose a novel infrared image destriping method, called Asymmetric Sampling Correction Network (ASCNet), that can effectively capture global column relationships and embed them into a U-shaped framework, providing comprehensive discriminative representation and seamless semantic connectivity. Our ASCNet consists of three core elements: Residual Haar Discrete Wavelet Transform (RHDWT), Pixel Shuffle (PS), and Column Non-uniformity Correction Module (CNCM). Specifically, RHDWT is a novel downsampler that employs double-branch modeling to effectively integrate stripe-directional prior knowledge and data-driven semantic interaction to enrich the feature representation. Observing the semantic patterns crosstalk of stripe noise, PS is introduced as an upsampler to prevent excessive apriori decoding and performing semantic-bias-free image reconstruction. After each sampling, CNCM captures the column relationships in long-range dependencies. By incorporating column, spatial, and self-dependence information, CNCM well establishes a global context to distinguish stripes from the scene’s vertical structures. Extensive experiments on synthetic data, real data, and infrared small target detection tasks demonstrate that the proposed method outperforms state-of-the-art single-image destriping methods both visually and quantitatively. Code is available at https://github.com/xdFai/ASCNet.
Fusion of infrared and visible images is a significant research area in image analysis and computer vision. The purpose of infrared and visible image fusion is to combine the complementary image information of the source images into a fused image. Thus, it is vital to efficiently represent the important image information of the source images and choose rational fusion rules. To achieve this aim, an image fusion method using multiscale directional nonlocal means (MDNLM) filter is proposed in this paper. The MDNLM combines the feature of preserving edge information by the nonlocal means filter with the capacity of capturing directional image information by the directional filter bank, which can effectively represent the intrinsic geometric structure of images. The MDNLM is a multiscale, multidirectional, and shift-invariant image decomposition method, and we use it to fuse infrared and visible images in this paper. First, the MDNLM is discussed and used to decompose the source images into approximation subbands and directional detail subbands. Then, the approximation and directional detail subbands are fused by a local neighborhood gradient weighted fusion rule and a local eighth-order correlation fusion rule, respectively. Finally, the fused image can be obtained through the inverse MDNLM. Comparison experiments have been performed on different image sets, and the results clearly demonstrate that the proposed method is superior to some conventional and recent proposed fusion methods in terms of the visual effects and objective evaluation.
The aim of multi-focus image fusion technology is to integrate different partially focused images into one all-focused image. To realize this goal, a new multi-focus image fusion method based on a guided filter is proposed and an efficient salient feature extraction method is presented in this paper. Furthermore, feature extraction is primarily the main objective of the present work. Based on salient feature extraction, the guided filter is first used to acquire the smoothing image containing the most sharpness regions. To obtain the initial fusion map, we compose a mixed focus measure by combining the variance of image intensities and the energy of the image gradient together. Then, the initial fusion map is further processed by a morphological filter to obtain a good reprocessed fusion map. Lastly, the final fusion map is determined via the reprocessed fusion map and is optimized by a guided filter. Experimental results demonstrate that the proposed method does markedly improve the fusion performance compared to previous fusion methods and can be competitive with or even outperform state-of-the-art fusion methods in terms of both subjective visual effects and objective quality metrics.
Convolutional neural networks have recently been used for multi-focus image fusion. However, some existing methods have resorted to adding Gaussian blur to focused images, to simulate defocus, thereby generating data (with ground-truth) for supervised learning. Moreover, they classify pixels as ‘focused’ or ‘defocused’, and use the classified results to construct the fusion weight maps. This then necessitates a series of post-processing steps. In this paper, we present an end-to-end learning approach for directly predicting the fully focused output image from multi-focus input image pairs. The suggested approach uses a CNN architecture trained to perform fusion, without the need for ground truth fused images. The CNN exploits the image structural similarity (SSIM) to calculate the loss, a metric that is widely accepted for fused image quality evaluation. What is more, we also use the standard deviation of a local window of the image to automatically estimate the importance of the source images in the final fused image when designing the loss function. Our network can accept images of variable sizes and hence, we are able to utilize real benchmark datasets, instead of simulated ones, to train our network. The model is a feed-forward, fully convolutional neural network that can process images of variable sizes during test time. Extensive evaluation on benchmark datasets show that our method outperforms, or is comparable with, existing state-of-the-art techniques on both objective and subjective benchmarks.