户外拍摄的视频图像经常会受到雾的影响, 被影响区域的图像往往清晰度差, 在进行视频稳像时不能正确地提取出特征点。如果采用去雾算法进行预处理, 也会造成特征点的误提取以及位置偏移等, 因此, 采用半逆法对图像进行分割, 分割出无雾区域以便在这些区域获得准确的特征点。由于半逆法采用固定阈值的方法, 无法满足不同图像的要求, 并且分割后的图像中存在多个孤立点, 因此, 采用自适应阈值和八邻域法对算法进行了改进。实验结果表明, 改进后的算法能够快速准确地分割无雾区域, 提高了特征点提取的准确率。

The outdoor video image is often affected by the fog. The affected area of the image is not clear and the feature points can’t be correctly detected during the video stabilization. If image dehazing algorithm is adopted to the pretreatment, which often leads to some false feature detection and positions offset etc. Therefore, the semi-inverse approach is proposed in this paper to segment the image and which can accurately detect the feature points in the absence of fog area. The fixed threshold selection of semi-inverse approach can’t meet the requirements of different images. Moreover, there are multiple isolated points in the image after segmentation. Therefore, the adaptive threshold and eight-neighborhood method are proposed in the paper to improve the algorithm. Experimental results show that the proposed algorithm can rapidly and accurately segment the image and get no fog area which can improve the accuracy of the feature detection.