Abstract: In foggy environment, the pictures taking from cameras have a poor visibility for human or device to identify the details. This paper introduces a novel approach to detect the haze and restore the image. The presented algorithm allows for a fast identification of hazy regions of a single image without expensive optimization or refinement procedures. By applying a semi-inverse operation to the original image, a semi-inverse image is produced. Comparing the pixels hue disparity between the original and its semi-inverse, the hazy regions are identified by setting a threshold value. Based on the optical model and the improved layer-based approach, the airlight constant can be estimated and several transmission images can be produced and then part of the layers are combined into one defogged image. The contrast stretching transformation approach is applied to the original defogged image to enhance its contrast.

Key words: haze detection, semi-inverse approach, improved layer-based approach, contrast stretching transformation approach, haze removing

摘要： 针对雾天环境下相机所拍摄图像能见度较差造成人眼或设备难以分辨图像细节的问题，提出一种新颖的单幅图像雾气检测和去除方法。在不经过图像优化及修复处理前提下，用半逆解法产生半逆解图像，通过阈值将转换至HSI 彩色空间的原图像和半逆解图像进行色度值对比，快速检测出图像雾气区域。依据光学模型估计出空气光常数等参数，用改进图层法作出多幅光传播图像，再将部分图层叠加成一幅图像。采用对比度拉伸变换法增加融合图像对比度，得到最终去雾图像。

关键词: 雾气检测, 逆解法, 改进图层法, 对比度拉伸变换法, 去雾