摘要：

针对图像去雾领域的暗通道先验算法存在光晕效应、颜色失真及对天空区域处理不佳等问题，提出一种改进暗通道先验和策略性融合的图像去雾算法。依据暗通道先验原理得到多尺寸最小值滤波窗口下的透射率和对应大气光值，设计基于结构相似性指标的调节和侧重因子用于多窗口透射率拟合。利用非线性规划模型和偏度理论计算全局大气光值，引入并调整置信度对天空区域的透射率进行快速补偿，结合大气散射模型恢复去雾图。将去雾图转换成HSV模型并对亮度V空间进行增强，同时对增强前后的图像进行策略性融合。实验结果表明，相比于现有的暗通道先验去雾算法，该算法的可见边增率、信息熵和平均梯度分别提升了79%~131%、3%~9%、39%~81%，具有较好的去雾效果，避免了光晕效应和颜色失真现象，适用于处理包含不同尺寸天空区域的雾图。

关键词: 图像去雾, 暗通道先验, 结构相似性, 非线性规划, 置信度, 策略性融合

Abstract:

To address the problems of halo effect, color distortion, and poor treatment of sky regions in the Dark Channel Prior(DCP) algorithm for image dehazing, a novel image dehazing algorithm using an improved DCP and strategic fusion is proposed. Based on the DCP principle, the transmission rate and corresponding atmospheric light value are obtained using a multi-scale minimum value filtering window. A design utilizing Structural Similarity(SSIM) index is employed to adjust and emphasize factors for multi-transmission rate fitting. A nonlinear optimization model and the skewness theory are utilized to calculate the global atmospheric light value, and confidence is introduced and adjusted to rapidly compensate for the transmission rate in sky regions. Through combination with an atmospheric scattering model, the dehazed image is restored. The dehazed image is then transformed into the HSV color space to enhance the brightness in the V channel, followed by a strategic enhancement fusion. Experimental results show that compared with the existing DCP dehazing algorithms, the proposed algorithm achieves significant improvements in visible edge increment rate, information entropy, and average gradient by 79%-131%, 3%-9%, and 39%-81%, respectively. This demonstrates its superior dehazing performance by avoiding halo effects and color distortion phenomena, and it is suitable for handling hazy images containing sky regions of different sizes.

Key words: image dehazing, Dark Channel Prior(DCP), Structural Similarity(SSIM), Non-Linear Programming(NLP), confidence, strategic fusion