摘要
矩不变量自从提出以来,在模式识别等方面得到了广泛的应用。由于受矩本身计算量的限制,很难实现实时计算,在高维空间中这一问题显得尤为突出。本文提出了一种建立在图像投影基础上的三维矩快速算法。该方法利用投影变换把三维空间的信息投影到一维空间,大大减少了运算量。利用该方法可以把三维矩计算的乘法运算复杂度从O(r^nN^n)降低到O(r^N),同时,加法运算复杂度也有很大程度的降低,并且,在运算过程中,不带来任何额外误差。本文还把这一方法推广到n维的情况,使得n维矩计算的乘法运算复杂度从O(r^rN^n)降低到O(rN)
Moment invariants have been widely used in pattern recognition and other fields since- they wen; proposed. Because of the complexity of the moments, it is difficult to implement real time computation, especially in high dimensional space. A new image-projection - based fast algorithm of 3D moments is presented in this paper. The information of 3D space is projected to 1D space, so the complexity of computation is greatly reduced. Using this algorithm, the complexity of multiplication is reduced from O(r3 N3) to O(rN) , and the complexity of addition is also reduced greatly. In addition, there is no extra error introduced in processing. The algorithm can be generalized to n-dimensional case, and the complexity of multiplication is reduced from O( rnNn) to O( rN) .
出处
《模式识别与人工智能》
EI
CSCD
北大核心
2001年第1期59-63,共5页
Pattern Recognition and Artificial Intelligence
