基于失真统计特征提取的图像尘雾清晰化算法

An Algorithm for Image Dust-Fog Sharpening Based on Distortion Statistical Feature Extraction

由于传统图像尘雾清晰化方法缺少图像纹理误差调整的过程,导致图像失真、对比度低以及图像整体泛白等问题,提出基于失真统计特征提取的图像尘雾清晰化算法。构建深度融合网络,对原始尘雾图像进行对比度增强、亮度修正与伽马校正等预处理,提升原始尘雾图像清晰度与对比度;综合考虑预处理后尘雾图像的灰度特征与纹理特征,利用模糊C均值法统计图像特征并进行聚类,融合参与特征提取的主要参数构建基于失真统计特征融合的图像特征提取模型,提取分类后的尘雾图像失真特征;训练深入学习网络,优化尘雾图像失真特征的权重与偏置项参数,输出尘雾清晰化的图像。仿真结果表明,上述算法进行尘雾图像清晰化还原后,相似度指数与峰值信噪比分别保持在0.84和34.21dB以上,显著优于对比算法,为提高尘雾图像处理效果提供了有利依据。

The traditional method lacks the process of adjusting the image texture error,resulting in image distortion,low contrast and overall whiteness.Therefore,an algorithm of dust-fog image sharpening based on distortion statistical feature extraction was proposed.The deep fusion network was constructed to enhance the contrast and correct the brightness and Gamma of original image,so as to improve the clarity and contrast of original dust-fog image.After comprehensively considering the gray-scale feature and texture feature of the pre-processed image,the fuzzy c-means method was used to calculate the image features and agglomerate them.Moreover,the main parameters involved in feature extraction were used to construct the model of image feature extraction based on the fusion of distortion statistical features.In addition,the dust-fog image distortion features after classification were extracted.Furthermore,the deep learning network was trained.Meanwhile,the weight and offset parameters of dust-fog image distortion features.Finally,the clear image was output.Simulation results show that the proposed algorithm keeps the similarity index and peak SNR above 0.84 and 34.21 db respectively after the dust-fog image restoration,which is significantly better than the comparison algorithm.Therefore,this algorithm provides a favorable basis for improving the effect of processing dust-fog image.