【摘要】 针对超像素分割算法需要人为设置初始超像素个数和目标边缘分割不精确等问题,提出一种自适应产生超像素个数的道路图像分割算法。该算法主要包含超像素的获得和超像素的合并两阶段。在超像素的获得阶段,首先通过计算图像区域数对应的图像颜色分量直方图峰值个数自动获得初始超像素个数,然后基于SLIC(simple linear iterative clustering)算法在图像过分割的基础上利用颜色分量最大差值对过分割超像素块进行欠分割检测与处理,实现超像素的精确分割。在超像素的合并阶段,通过融合超像素颜色和纹理特征建立超像素间相似度信息表,最后在结合空间位置相邻性的基础上实现超像素的合并。实验在自动驾驶场景评测数据集KITTI上对算法进行验证和测试。结果表明,提出的算法与其他道路图像分割算法相比,在总体精度、平均召回率以及F1值3个指标上均有较好的效果。更多还原

【Abstract】 Aiming at the problem that the superpixel algorithm needs to artificially set the initial number of superpixels and the object edge segmentation are inaccurate,a road image segmentation algorithm using generate adaptively the number of superpixels is proposed. The algorithm mainly includes two stages of obtaining superpixels and merging superpixels. In the obtaining stage of the superpixels,the initial number of superpixel is automatically obtained by calculating the peak number of the image color component histogram corresponding to the number of regions of the image. The image are over-segmented using（ SLIC simple linear iterative clustering）,then over-segmentation detection and processing are performed based on the maximum difference of the color component,which performs accurate segmentation of the superpixels. In the merging stage of superpixels,a similarity table between superpixels is established by merging superpixels’ color and texture features,and finally the superpixel merging is achieved on the basis of considering the spatial location neighboring. The experiment validates and tests the algorithm on the automatic driving scenario evaluation KITTI data set. The results show that compared with the other road image segmentation algorithms,it has better results in terms of overall accuracy,average recall,and F₁ value.更多还原