CAJ | 학술논문

根据人头特征,提出了一种基于网格和密度的聚类算法.该算法将图像分成网格,然后逐行计算网格的密度,碰到符合密度要求的网格时,算法转为纵向计算网格的密度,记录下纵向符合密度要求的网格数量,以此判断是否存在人头以及计算人头的参数.该算法结合了网格聚类的低时空复杂度和密度聚类的良好抗噪性的特点.实验证明该算法速度比Hough变换快两个数量级,而且所需存储空间小.
근거인두특정,제출료일충기우망격화밀도적취류산법.해산법장도상분성망격,연후축행계산망격적밀도,팽도부합밀도요구적망격시,산법전위종향계산망격적밀도,기록하종향부합밀도요구적망격수량,이차판단시부존재인두이급계산인두적삼수.해산법결합료망격취류적저시공복잡도화밀도취류적량호항조성적특점.실험증명해산법속도비Hough변환쾌량개수량급,이차소수존저공간소.
According to the features of human head,the paper puts forward a new clustering algorithm based on grid and density.In the algorithm,image is divided into grids,then calculates the density of every grid line by line.When coming across a grid that meets the density requirement,algorithm turns to calculate the density of grid vertically,and records the number of vertical grids that meets the density requirement.Then judge whether there are human heads and calculate the parameters of head.This algorithm has the merit of grid-based clustering which is low-complexity in time and space, and has the merit of density-based clustering which is good noise immunity.The experimental results show that it is two orders magnitude faster than Hough transform, and requires small storage space.