| 基于自适应多特征融合的红外图像增强算法 |
| 投稿时间：2024-01-26 修订日期：2024-03-06 点此下载全文 |
| 引用本文： |
| 摘要点击次数: 58 |
| 全文下载次数: 0 |
|
| 基金项目:国家自然科学基金(No.621713600)、陕西省科技厅重点研发计划(2022GY-110)、国家重点研发计划(2022YFF0604900)2022年度陕西高校青年创新团队项目；山东省智慧交通重点实验室(筹)；2023年陕西省高校工程研究中心。 |
|
| 中文摘要:针对红外图像的纹理不清晰，亮度低、高噪声的问题，本文提出了一种自适应多特征融合的红外图像增强算法。首先，通过自动线性映射的方法将14位的红外图像进行有效特征提取得到了16位图像，提升了图像可视化效果。其次，引入广义反锐化掩模算法(A Generalized Unsharp Masking Algorithm, GUM)与带色彩恢复的多尺度视网膜增强算法(Multi-Scale Retinex with Color Restoration, MSRCR）联合处理的方法，获得图像不同尺度的有效信息，提升了图像的对比度。最后，设计了自适应权重图并结合图像金字塔结构的特性，对不同特征层进行有效信息的互补融合，提升了图像亮度，丰富了图像的纹理信息。实验结果表明，此算法有效提升了红外图像的对比度和视觉效果，相对于现有的几种算法，平均梯度约提升0.6%，峰值信噪比约提升10%，图像的边缘信息有效率约提升11%，图像的清晰度约提升10%。 |
| 中文关键词:特征提取 权重图 金字塔 多尺度融合 |
| |
| Infrared Image Enhancement Algorithm Based on Adaptive Multi- Feature Fusion |
|
|
| Abstract:Aiming at the existing infrared image enhancement algorithms with unclear texture，low brightness and high noise, an enhancement algorithm capable of adaptively fusing multiple features in the processed infrared images is proposed. First of all, an automatic linear mapping method is designed to extract feature 14bit infrared image to 16bit image for normal display to improve the image visualization. Secondly, for the problem that the infrared image is low contrast, a joint processing is designed introduce the Generalized Unsharp Masking Algorithm (GUM) and Multi Scale Retina with Color Restoration (MSRCR) algorithms to obtain effective information from images at different scales to improve the image contrast. Finally, an adaptive weight map is designed according to the exposure degree of image and combined with features of the image pyramid structure to achieve complementary fusion of effective information from different feature layers, and improve image brightness and enhance image texture information. The experimental results show that this algorithm effectively improves the contrast and visual effect of infrared images, compared to existing algorithms, the average gradient is increased by 0.6%, the PSNR is increased by 10%, the effective efficiency of edge information is increased by 11%, and the clarity of the image is increased by 10%. |
| keywords:feature extraction weight map pyramid multiscale fusion |
| HTML 查看/发表评论 下载PDF阅读器 |