摘要
针对在线K-均值聚类法初始化混合高斯模型(KGMM)在运行时间、空间复杂度、噪声等方面存在的缺陷,提出了基于KGMM改进的检测方法,采用加入方差因子的C-均值聚类准则来初始化混合高斯模型,有效解决了可能出现的某一像素值属于不同分布类从而概率不同的问题,提高了检测的灵活性;改进了高斯匹配准则,提高了检测算法的准确性;对每个像素点间隔地建立混合高斯分布,减少了高斯模型个数,节省了存储空间,提高了算法的运行速度。实验结果表明改进的检测算法检测效果更理想。
The online K-means clustering method for initialization Gaussian mixture model （KGMM） with respect to run time, space complexity and noise have some disadvantages, this paper proposed an improved method of detection based on KGMM, added the variance factor to the C-means clustering criterion to initialize Gaussian mixture model. It effectively solved the problem that a pixel value may belong to different distribution classes driving different probabilities, and improved the flexibility of detection ; improved the matching criterion of Gaussian model and increased the accuracy of the detection algorithm; established mixed Gaussian distribution for every other pixel point, it reduced the amount of Gaussian model, saved storage space, and reduced the running time of the algorithm. The experimental results show that the effect of the improved detection algorithm is more ideal.
出处
《计算机应用研究》
CSCD
北大核心
2012年第8期3189-3191,共3页
Application Research of Computers
基金
国家自然科学基金资助项目(50877010)
