CAJ | 학술논문

根据核函数的不同特点,分析得出均值移动算法的步长可能为牛顿步长、高斯-牛顿步长或一种非线性步长.根据均值移动密度函数泰勒展开形式的不同,讨论均值移动点每步在运动方向上的密度递增范围和密度极大值点的位置.上述结论对于提高均值移动算法的收敛速度有指导作用.
근거핵함수적불동특점,분석득출균치이동산법적보장가능위우돈보장、고사-우돈보장혹일충비선성보장.근거균치이동밀도함수태륵전개형식적불동,토론균치이동점매보재운동방향상적밀도체증범위화밀도겁대치점적위치.상술결론대우제고균치이동산법적수렴속도유지도작용.
According to the different characteristics of kernel functions, this paper concludes that the mean-shift step may be Newton-step, gauss Newton-step or nonlinear-step. The bound of the density increasing and the position of density maximum at every step are discussed according to the different Taylor series of the mean-shift density functions. The convergence speed can be increased based on such conclusions.