Abstract: In order to establish a target model to describe image structure characteristics in the field of computer vision, a quad-tree partition model based on density estimation is proposed. The local density estimation model is established by using quad-tree partition method in the target image block division, according to statistical features of pixel values and scales of each block. Analysis of experimental results shows that the method not only describes structural characteristics which the global density estimation model cannot do, but also describes details of the target distinctly with less number of blocks. At the same time, algorithm based on the proposed model simplifies calculation process of density estimation and has obvious advantages of target tracking applications over other algorithms.

Key words: block tracking, quad-tree, density estimation, nonparametric estimation

摘要： 针对计算机视觉领域中，建立能够描述图像结构特性的目标模型问题，提出了一种基于四叉树分块的密度估计模型。采用四叉树分块方法对目标图像块进行划分，根据各分块的尺度和像素值统计特性建立局部的密度估计模型。实验结果和分析表明，基于四叉树分块的局部密度估计模型解决了全局密度估计模型无法描述结构特性的问题，而且用较少的分块数量有针对性地描述了目标的细节，同时简化了密度估计算法的计算过程，在目标跟踪应用中较其他算法表现出较优的性能。

关键词: 分块跟踪, 四叉树, 密度估计, 非参数估计