【摘要】 夜晚由于人工光源的存在,常规的基于非物理模型的白天去雾算法不适用夜晚去雾。夜晚采集到的图像具有偏色、整体亮度较低和光照不均匀等特点。针对光照不均匀问题,利用传递函数为高斯低通型的同态滤波进行滤波处理,以校正图像光照;针对光晕问题,经光照校正的图像再利用Retinex算法去除光源附近的光晕;针对夜晚图像整体亮度较低问题,采用CLAHE算法提高夜晚图像暗区域的对比度;针对色偏问题,利用白平衡算法进行颜色校正,以符合人眼的视觉特性。通过对夜晚有雾图像进行实验,以平均梯度、信息熵、对比度和算法时间为客观指标,与传统Retinex算法、白天和夜晚图像去雾算法比较,结果表明该算法在以上客观指标中总体表现最优,验证了该算法的有效性。更多还原

【Abstract】 Due to the existence of artificial light at night, the conventional daytime dehazing algorithms based on non physical model are not suitable for nighttime dehazing. Nighttime haze images have the characteristics of uneven illumination, low overall brightness and color deviation. For uneven illumination, transfer function was used to process the homomorphic filtering of Gaussian lowpass filtering to correct the image illumination. For the halo, the Retinex algorithm was used to remove the halo of the illumination-corrected image near light sources. For low overall brightness of nighttime images, CLAHE algorithm was applied to improve the contrast ratio of dark areas on nighttime images. For color cast, the white balance method was adopted to correct colors to meet the visual characteristics of human eyes. The experiment of nighttime haze images was performed to compare such objective indicators as average gradient, information entropy, contrast ratio, and time complexity with traditional Retinex algorithm, as well as nighttime and daytime image dehazing algorithms. Indicators of the algorithm proposed in this paper are the best, which verifies its effectiveness.更多还原