【摘要】 针对SNIC(Superpixels and Polygons using Simple Non-Iterative Clustering)算法在进行距离度量时空间距离和颜色距离采用的距离比例系数恒定不变，而忽视了像素点的局部特点，提出了一种基于自适应距离度量的改进超像素算法。算法引入局部二值模式（Local Binary Pattern,LBP），计算并保留所有像素点的LBP值，在进行种子点更新时保存标签内像素LBP的平均值，并通过像素点与种子点的LBP值自适应地计算距离系数，采用新的距离公式计算距离并迭代更新。在C++环境下分别对所提算法与四种对比算法进行测试，在超像素个数均为600的情况下，所提算法在BSD500数据集上与经典的SNIC算法相比，在边界召回率方面提升3.9%，欠分割错误率降低1.1%。实验结果表明，所提算法能有效提升超像素分割质量。更多还原

【Abstract】 Aiming at the constant distance proportionality of spatial distance and color distance in SNIC(Superpixels and Polygons using Simple Non-Iterative Clustering) algorithm,while ignoring the local characteristics of pixels,an improved superpixel algorithm based on adaptive distance measurement was proposed. The Local Binary Pattern(LBP) is introduced. The LBP values of all pixels are calculated and retained,the average value of LBP of pixels in the label is saved during seed point update,and the distance coefficient is calculated by the adaptive LBP values of pixels and seed points. A new distance formula is used to calculate the distance and update iteratively. In the C++ environment,the proposed algorithm and the four comparison algorithms are tested respectively. When the number of superpixels is600,the proposed algorithm has improved boundary recall rate compared with the classic SNIC algorithm on the BSD500 data set 3.9%,and the under-segmentation error rate is reduced by 1.1%. The experimental results show that the proposed algorithm can effectively improve the quality of superpixel segmentation.更多还原