摘要
快速准确地估计全局运动参数,是模型基图像编码中需要解决的关键问题之一。在优化最小二乘算法中最常用的是Ｌ－Ｍ方法,但其为得到一个合适的阻尼因子需要多次矩阵求逆,计算量颇大,很难满足实时性要求,为克服此缺陷,本文在估计算法中引入了螺线方法,提出了螺线方法的两种改进形式－中点螺线方法和角度螺线方法。根据实际系统的实时性要求,还给出了一种灵活调整搜索时间的方法。并对传统的Ｌ－Ｍ方法和改进的两种螺线方法进行了实验对比。结果表明:在保证估计性能相当的情况下,这两种算法显著地提高了估计速度。
How to estimate the motion parameters fast and accurately is one of the problems required to be solved urgently in model-based image coding. The most often used method in optimal least-squares estimation algorithms is Levenberg-Marquart method, which computation complexity is burdensome due to many matrix inversions to find a moderate damping factor. In order to overcome its disadvantages, a method named SPIRAL was imported. Two modified method of Spiral was proposed as midpoint Spiral method and angle Spiral method. A method was also presented to adjust searching time flexibly according to the individual real-time requirements of applications. Experiment results were presented for synthetic data and real long image sequences. Compared with conventional L-M method, the results demonstrates that two modified algorithms excel distinctly in the estimating time per frame with the same estimation performance.
出处
《信号处理》
CSCD
2002年第3期249-253,共5页
Journal of Signal Processing
基金
国家自然科学基金资助项目(60072013)
