For a natural scene with nonuniform environment light, the captured visible images are always under- or over-exposed because of the limited dynamic range of digital imaging devices. Multi-exposure image fusion (MEF) is a mainstream and effective solution. For a local region that has friendly visual effect in one exposure setting but extremely bad-exposed in another, most existing MEF methods have the ability to transfer the scene detail information to the fused images. However, they will be affected by the over-high or -low light inevitably thus resulting in local visibility reduction. To address this issue, we propose an adaptive clarity evaluation-guided network with illumination correction for MEF in a coarse-to-fine manner, which is termed as ACE-MEF. To be specific, our ACE-MEF is mainly composed of two modules: clarity preservation network (CPN) and illumination adjustment network (IAN). Based on the adaptive clarity evaluation, CPN could be trained to coarsely preserve the environment light and texture details of the clearer regions in source images. Therefore, the need for labeled reference images that are time-consuming to obtain could be mitigated. By measuring the parameter maps of gamma function, IAN is able to refine and correct the local bad-exposed regions so that more details could be further revealed. Extensive experiments demonstrate that our method outperforms multiple state-of-the-art algorithms qualitatively and quantitatively.

Dear Editor, This letter is concerned with dealing with the great discrepancy between near-infrared (NIR) and visible (VS) image fusion via color distribution preserved generative adversarial network (CDP-GAN). Different from the global discriminator in prior GAN, conflict of preserving NIR details and VS color is resolved by introducing an attention guidance mechanism into the discriminator. Moreover, perceptual loss with adaptive weights increases quality of high-frequency features and helps to eliminate noise appeared in VS image. Finally, experiments are given to validate the proposed method.

In this paper, we reveal and rectify two underlying factors that hinder the performance of previous PatchMatch-based Multi-View Stereo (MVS) methods in texture-less regions. First, previous methods approximate the desire pixel-wise similarity measurement with the patch-wise counterpart, which tends to be misled by ambiguous matches in texture-less areas. Therefore, we propose a novel similarity measurement that takes into account the relation between neighboring pixels and the central one. Second, the projection formula adopted by previous methods mistakes invisible points behind the optical center as visible, which leads to wrong results in view selection and depth maps. We not only rectify the formula, but also propose a geometric restriction that eliminates wrong results by shrinking the solution space of the hypothetical normal vector. These two improved aspects are necessary components of the traditional MVS pipeline and can therefore be easily applied to other PatchMatch-based MVS methods. Experiment results show that the proposed approach can effectively suppress the noise in depth maps while preserving fine details. In terms of F ₁ -score, it achieves state-of-the-art performance on ETH3D high-resolution MVS benchmark.

The cooled mid-wave infrared biomimetic compound eye camera has wide range of applications, such as industrial inspection, military project, and security. Due to the low resolution of individual eyes and the large field view of the imaging system, existing motion target enhancement and detection algorithms cannot effectively detect all potential targets. To address this issue, we propose an improved elementary motion detector model that combines a double-layer ON_OFF channel and a cross-type computational architecture, which is able to suppress a stationary background and enhance moving targets. In order to further reduce the missed detection, we designed the spatial and temporal consistency detection methods of the compound eye structure, which further improved the accuracy and stability of the detection results. The experimental results show that our method can fully utilize the features of the image and can be applied to the enhancement and detection of moving objects in complex scenes, and the detection efficiency is significantly improved.

Traditional 2D imaging technologies are limited by the need for a large field of view and their sensitivity to small target motion. Inspired by the characteristics of insect compound eye structure, we propose an infrared bionic compound eye camera based on a small lens array. The camera is composed of 61 small lens arrays mounted on a curved spherical shell and a relay optical system. The imaging device is a high-performance cooled mid-wave infrared detector. This is an innovative design for an infrared biomimetic compound eye camera system that provides a wide field of view and all-day detection capability. Aimed to meet the specified requirements. The optical system achieves a 100% cold-membrane match between the infrared optical system and the cooled detector, and the relay optical system optimizes the large-field aberration by introducing a higher-order aspheric surface and modifying the geometric surface of the lenses. Our entire system enables an observation field angle of $$108^\circ \times 108^\circ$$108 ∘ × 108 ∘ . The experiments showed that the image quality of the system is high, each ommatidium was effective within the imaging range of the compound eye camera, resulting in an improved signal-to-noise ratio in various scenes.

Building segmentation for Unmanned Aerial Vehicle (UAV) imagery usually requires pixel-level labels, which are time-consuming and expensive to collect. Weakly supervised semantic segmentation methods for image-level labeling have recently achieved promising performance in natural scenes, but there have been few studies on UAV remote sensing imagery. In this paper, we propose a reliable label-supervised pixel attention mechanism for building segmentation in UAV imagery. Our method is based on the class activation map. However, classification networks tend to capture discriminative parts of the object and are insensitive to over-activation; therefore, class activation maps cannot directly guide segmentation network training. To overcome these challenges, we first design a Pixel Attention Module that captures rich contextual relationships, which can further mine more discriminative regions, in order to obtain a modified class activation map. Then, we use the initial seeds generated by the classification network to synthesize reliable labels. Finally, we design a reliable label loss, which is defined as the sum of the pixel-level differences between the reliable labels and the modified class activation map. Notably, the reliable label loss can handle over-activation. The preceding steps can significantly improve the quality of the pseudo-labels. Experiments on our home-made UAV data set indicate that our method can achieve 88.8% mIoU on the test set, outperforming previous state-of-the-art weakly supervised methods.