质心迭代图像跟踪算法

针对均值漂移跟踪算法对图像进行核函数加权处理的不合理性,且在图像跟踪中存在偏差的问题,提出了一种基于最大后验概率指标的质心迭代跟踪算法.首先分析了最大后验概率指标的计算特性,并指出该指标可以计算出每个像素对相似度的贡献值.以此为基础,提出了一种非参数非核的质心迭代图像跟踪算法,即将每个像素的相似度贡献作为密度,候选区域的相似度作为区域质量,通过计算初始候选区域的质心并经反复迭代,从而获得目标位置.理论分析和实验表明,所提算法无需核函数,迭代计算无需指数运算,降低了计算复杂度,同时利用了最大后验概率指标对背景的抑制作用,可大大提高跟踪的准确性.

Centroid Iteration Image Tracking Algorithm

In order to avoid the unreasonable kernel weight imposed on images by mean-shift tracking algorithm and the tracking bias, a maximum posterior probability measure based centroid iteration tracking algorithm is developed. The computation property of maximum posterior probability similarity measure is first analyzed, based on which it is indicated that the contribution of each pixel to the similarity value can be calculated by this measure. On the basis of it, a non-kernel and non-parametric centroid iteration image tracking algorithm is proposed, which takes the similarity contribution of each pixel as density and the similarity value of the candidate as mass, and obtains the target region by moving iteratively to the next centroid from the initial target region centroid. Theoretical analysis and experimental results show that the new algorithm is nonkernel and needs no extraction computation, which reduces the computational complexity. Meanwhile, utilizing the depressing capability of maximum posterior probability measure on the background, the tracking precision can be greatly improved.