多尺度融合和细节提升的水下图像增强
Underwater Image Enhancement with Multi-scale Fusion and Detail Enhancement
李华昆 1赵磊 1李恒 1王海瑞1
作者信息
- 1. 昆明理工大学信息工程与自动化学院,昆明 650500
- 折叠
摘要
光在水中传播时,由于水的吸收和散射,在水下环境拍摄的图像常常会出现色偏、对比度低和细节丢失等问题.为了恢复图像的可视性和清晰度,提出一种多尺度融合和细节提升的水下图像增强算法.针对水下图像存在的偏色问题,提出一种基于通道补偿的颜色校正算法;为提高水下图像的亮度和对比度,对颜色校正图像的RGB三通道进行直方图拉伸改善亮度和全局对比度,再使用限制对比度的自适应直方图均衡化(Contrast limited adaptive histogram equalization,CLAHE)算法得到改善的局部对比度增强图像,然后对全局对比度增强图像和局部对比度增强图像进行多尺度融合;为提升图像细节,采用改进的Sobel算子提取图像边缘细节,添加到融合图像中,得到最终增强图像.最后通过与一些常用的水下图像处理方法进行比较,从主观视觉效果和客观评价两方面,证明本文所提算法取得较好效果.
Abstract
Due to the absorption and scattering of water when light travels through water,images taken in underwater environment often have problems such as color deviation,low contrast and loss of detail.An underwater image enhancement algorithm with multi-scale fusion and detail enhancement is proposed to restore the visibility and sharpness of the image.A color correction algorithm based on channel compensation is proposed to solve the color deviation problem of an underwater image.In order to improve the brightness and contrast of the underwater image,the histogram of three channels of the RGB color space of a color correction image is stretched to improve its brightness and global contrast.Then the contrast limited adaptive histogram equalization is used to improve the local contrast of the image.The global contrast enhancement image and the local contrast enhancement image are fused at multiple scales.For the detail loss caused by forward scattering,the improved Sobel operator is used to extract the image's edge details and add them to the fusion image to obtain the finally enhanced image.Finally,compared with some commonly used underwater image processing methods,the underwater image enhancement algorithm is effective for both subjective visual effect and objective evaluation.
关键词
图像增强/颜色校正/对比度增强/多尺度融合/细节提升Key words
image enhancement/color correction/contrast enhancement/multi-scale fusion/detail improvement引用本文复制引用
基金项目
国家自然科学基金(61863016)
地区科学基金(61263023)
出版年
2024
国家科技期刊平台
NSTL
万方数据