局部最近邻密度和颜色特征加权的超像素生成  

Superpixel Generation Based on Local Nearest Neighbor Density and Color Feature Weighting

作　　者：徐新黎[1] 邢少恒 王凯栋 许营坤 管秋[1] 王万良[1] XU Xin-li;XING Shao-heng;WANG Kai-dong;XU Ying-kun;GUAN Qiu;WANG Wan-liang(College of Computer Science and Technology,Zhejiang University of Technology,Hangzhou 310023,China)

机构地区：[1]浙江工业大学计算机科学与技术学院,杭州310023

出　　处：《小型微型计算机系统》2023年第1期117-123,共7页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61873240)资助;浙江省公益科技计划项目(LGG20F020017)资助;浙江省自然科学基金项目(LY20F020029,LY21F020027)资助。

摘　　要：作为计算机视觉领域的重要预处理步骤,超像素生成算法近年来受到广泛关注与研究.为了快速、高效地生成高质量的超像素,提出一种基于局部最近邻密度和颜色特征加权的超像素生成算法(NDPCS).算法分为两个阶段:1)结合图像中像素点最近邻居信息,计算各像素点的局部密度和局部密度最大值点决策值,选择拥有大决策值的像素点作为聚类中心,并根据颜色特征加权距离归类其他像素点,生成初始超像素;2)采用启发式合并策略,在保留边缘贴合度的前提下合并过小和孤立的初始超像素,保证超像素的连通性和一致性.实验在Berkeley数据集BSDS500上进行验证,本文所提方法在边缘召回率、欠分割误差和可达分割精度这些通用的评价指标上表现优良,可以为任意彩色图像快速生成高质量的超像素.As an important preprocessing step in the field of computer vision, the algorithms of superpixel generation have received extensive attention and research in recent years.In order to generate high quality superpixels quickly and efficiently, a method of superpixel generation based on local nearest neighbor density and color feature weighting was proposed(NDPCS).It consists of two stages.In the first stage, combining the information of K nearest neighbors of each pixel, the density of local neighborhood of the pixels and decision value of local maximum points are calculated, and then pixels with big decision value are selected as the clustering centers and other pixels are clustered by color feature weighting distance to clustering centers to generate the initial superpixels.In the second stage, a heuristic merging strategy is adopted to merge the initial superpixels that are too small and isolated under the premise of retaining the edge fit degree to ensure the connectivity and consistency of the superpixel.Experiments are carried out on the Berkeley dataset BSDS500.The proposed method performs well on the general evaluation indexes such as boundary recall, undersegmentation error and achievable segmentation accuracy, and it can quickly generate high-quality superpixels for any colour image.

关 键 词：超像素分割 密度峰值 像素聚类 局部密度 图像处理 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]