基于多重先验的无监督学习红外图像增强算法
Infrared Image Enhancement Algorithm Based on Unsupervised Learning of Multiple Priors
杨家全 1李邦源 2丁贞煜 3马文龙 4汪航 4孙宏滨3
作者信息
- 1. 西安交通大学人工智能学院,陕西 西安 710049;云南电网有限责任公司电力科学研究院,云南 昆明 650217
- 2. 云南电网有限责任公司玉溪供电局,云南 玉溪 653100
- 3. 西安交通大学人工智能学院,陕西 西安 710049
- 4. 西安交通大学微电子学院,陕西 西安 710049
- 折叠
摘要
随着红外成像技术的广泛应用,人们对红外成像质量的要求不断提高.由于红外传感成像原理的限制,红外图像普遍存在对比度不高、缺乏细节纹理等问题.与此同时,常用的传统红外图像增强算法在提升图像对比度的同时容易引入较高的噪声,导致增强效果有限.因此,本文提出一种基于高斯-拉普拉斯金字塔、CLAHE、原始图像等多重先验知识的无监督红外图像增强算法,并通过实验与各种经典先验方法进行了对比.实验证明,本算法效果相较于传统算法有较为明显的提升,可以完全自适应地实现对比度拉伸、去除噪声等红外图像增强操作,并且对目标检测等下游任务带来了显著的精度提升.
Abstract
With the wide application of infrared imaging technology,people's requirements for infrared imaging quality are also increasing.Due to the limitation of infrared imaging principle,infrared images generally have the problems of low contrast and lack of detail texture.Meanwhile,traditional infrared image enhancement algorithms introduce high noise while improving image contrast.Therefore,this paper proposes an unsupervised infrared image enhancement algorithm based on multiple prior knowledge such as Gaussian-Laplacian pyramid,CLAHE and original image,and compares it with other classical prior methods through experiments.Experiments show that the effect of the proposed algorithm is significantly improved compared with the traditional algorithms.It can fully adaptively realize infrared image enhancement operations such as contrast stretching and noise removal,and bring significant improvement to downstream tasks such as target detection.
关键词
红外图像/自适应对比度增强/无监督学习/高斯-拉普拉斯金字塔/CLAHEKey words
Infrared Image/Adaptive Contrast Enhancement/Unsupervised Learning/Gaussian-Laplacian Pyramid/CLAHE引用本文复制引用
基金项目
云南电网科技项目(YNKJXM20220023)
出版年
2024
NETL
NSTL
万方数据