The key issue of multi-sensor image fusion is how to accurately extract and fuse the high-quality pixels or coefficients of source images. Nevertheless, the so-called high-quality is an uncertain or fuzzy definition, which is very suitable for fuzzy theory to address this problem. By the integration of stationary wavelet transform (SWT) and fuzzy sets, this paper proposes a new multi-focus image fusion scheme, which can merge the important features of different source images into a fused image. First, the source images are decomposed by SWT to get a set of sub-images with different detailed features. Second, the Gaussian membership function (GMF) is utilized to get the fuzzy sets of sub-images data. Third, the local spatial frequency (LSF) is employed to extract the local features of the sub-images by the generated fuzzy sets. At last, the fusion rule is designed based on consistency verification to fuse the sub-images according to the LSF of fuzzy sets, and then inverse SWT (ISWT) is implemented to reconstruct the fused image. The experimental and contrastive results of different image sets show that the proposed method is an effective multi-focus image fusion scheme which can achieve better fusion effect than other methods.

Because thermal infrared (TIR) images are not affected by light and foggy environments, which are widely used in various night traffic scenarios. Especially, thermal infrared images also play an important role in autonomous vehicles. However, low contrast and lack of chromaticity have always been their problems. Image colorization is a vital technique to improve the quality of TIR images, which is beneficial to human interpretation and downstream tasks. Despite thermal infrared image colorization methods have been rapidly improved, the detail blurriness and color distortion in colorized images remain under-addressed. Mostly because these methods cannot effectively extract the ambiguous feature information of TIR images. Hence, we propose a large kernel (LK) U-Net and Attention_U-Net-Transformer (ViT-Based) based generative adversarial network. An LK_U-Net is designed to extract the feature of TIR images. Then, a branch structure composed of Attention_U-Net and ViT-Based can provide the network with semantic information from different perspectives to decode features. In addition, a composite loss function is employed to ensure the network generates a high-quality colorized image. The proposed method is evaluated on KAIST and IRVI datasets. Experimental results demonstrate the superiority of the proposed LKAT-GAN over other methods for the task of thermal infrared image colorization. The code is available at.

Computational imaging provides comprehensive and reliable information about human tissue for medical diagnosis and treatment, with medical image fusion as one of the most important technologies in the field. Empirical mode decomposition (EMD), a promising model for image processing, has been used for image fusion in some methods. However, the varying number of decomposed layers leads to problems using EMD for image fusion. In this work, we propose a fusion method for medical images incorporating L2-norm-based features, a match/salience/fuzzy-weighted measure, and the two-dimensional Littlewood-Paley empirical wavelet transform (2D LPEWT) as new version of EMD. We first decompose medical images with LPEWT to obtain the residual component (residue) and detailed sub-images (intrinsic mode functions, IMFs). Then we extract the regional features of residue with an L2-norm-based model to fuse the residue while simultaneously fusing IMFs using a method combining a fuzzy membership function with a match/salience measurement. Finally, we reconstruct the comprehensive image by applying inverse LPEWT to the fused residue and IMFs. We evaluated our method using a frequently-used dataset of brain images. The results show that our proposed method is more effective than conventional methods by fusing more information into the final images. We also show a feasible scheme for applying EMD to image fusion.

Medical image fusion combines the multiple features of human tissue from different source images, which will be conducive to clinical diagnosis. Because of the human visual perception and mechanisms of medical imaging, abundant fuzzy features exist in medical image fusion with remarkable influence. To address this issue, we introduce the similarity measure of fuzzy set theory into medical image fusion to abstract and measure these fuzzy features. In this study, we present a new similarity measure of intuitionistic fuzzy set theory to represent image features and present a lightweight medical image fusion technique on the basis of this new measure. First, the new similarity measure is proposed according to geometric modeling technique and verified by mathematical justification. Second, a set of high frequency sub-images and a low frequency sub-image are produced by Laplacian pyramid decomposition to separate different image features. Third, our proposed similarity measure is utilized to describe and extract the detailed image features. Finally, two fusion rules and inverse Laplacian pyramid decomposition are sequentially used to get the fused image. This research shows that the similarity measure of fuzzy set theory can get excellent performance in medical image fusion. Experiments reveal that our similarity measure and medical image fusion method both have the superior performance to most of the existing methods.