摘要
针对现有图像去雾算法在天空或高亮区域透射率估计不准确的问题,且复原图像色彩失真以及细节丢失严重等情况,提出了一种基于线性变换的自适应透射率去雾算法。首先,将输入图像转换至YCbCr空间提取亮度分量,并构造反S型函数对其进行尺度压缩,以此减弱高亮像素的影响;然后,利用线性变换模型对压缩后的亮度分量进行增强处理,使用高斯函数对亮度分量进行卷积操作得到自适应控制参数;结合线性变换模型和自适应控制参数逼近无雾图像最小颜色通道操作,进而得到精确的透射率估计值;最后,利用大气散射模型和局部大气光值逆向求解出复原图像。在实验验证中,采用可见边、平均梯度、饱和像素点及结构相似性作为客观评价指标。客观数据表明,所提算法的各项指标均取得优势。在主观效果方面,所提算法可以准确估计出透射率,有效去除图像雾气干扰并改善天空或明亮区域色彩失真的现象,提高图像可视度,复原出更多细节和边缘信息。
In order to solve the deficiencies of existing image defogging algorithms,such as inaccurate estimation of the transmission in the sky or bright areas,image color distortion and serious loss of details,an adaptive transmission defogging algorithm based on linear transformation was proposed.Firstly,the input image was converted to the YCbCr space to extract the brightness component,and an anti-S type function was constructed to reduce the influence of highlighted pixels.Then a linear transformation model is used to enhance the compressed luminance component.In order to obtain an adaptive control parameter,a Gaussian function was adopted to convolve the luminance component.The minimum color channel of the fog-free image was approximated by combining linear transformation model and the adaptive control parameter,and further an accurate estimate of the transmission was obtained.Finally,the restored image was acquired by using the atmospheric scattering model and local atmospheric light in reverse.In the experimental verification part,visible edges,average gradients,saturated pixels,and structural similarity were used as objective evaluation indicators.The objective data illustrated that all indicators of the proposed method achieves better performance.In terms of subjective effects,the proposed algorithm is superior to several existing defogging algorithms.It can accurately estimate the transmission,effectively remove image fog interference and improve the color distortion of the sky or bright areas,improve image visibility,and restore more details such as image edges.
作者
杨燕
姜沛沛
岳辉
YANG Yan;JIANG Peipei;YUE Hui(School of Electronic and Info.Eng.,Lanzhou Jiao Tong Univ.,Lanzhou 730070,China)
出处
《工程科学与技术》
EI
CAS
CSCD
北大核心
2020年第5期194-200,共7页
Advanced Engineering Sciences
基金
国家自然科学基金项目(61561030)
甘肃省财政厅基本科研业务费基金项目(214138)
兰州交通大学教改基金项目(160012)。
关键词
图像处理
反S型函数
线性变换
高斯函数
自适应参数
image processing
anti-S type function
linear transformation
Gaussian function
adaptive parameter
