摘要 |
PROBLEM TO BE SOLVED: To quickly calculate a regression coefficient by approximating arithmetic for approximating conventional statistic calculation. SOLUTION: A linear regression arithmetic unit 101 obtains an estimated value of a criterion variable from distributions of two data sequences of an explanatory variable and the criterion variable. A covariance approximation calculation part 113 performs approximating arithmetic of a covariance of signals 1 and 2. A variance approximation calculation part 114 performs approximating arithmetic of a variance of the signal 1. A regression line is given by y=b(x-Mx)+My, and a regression coefficient is equal to Vxy/Vxx wherein Vxy is a covariance value and Vxx is a variance value. In an estimated value arithmetic part 115, a subtraction part 121 obtains (x-Mx), and a division part 122 calculates an approximate value of the regression coefficient b by using an approximate value of the covariance value Vxy and that of the variance value Vxx. A multiplication part 123 multiplies an arithmetic result of the subtraction part 121 and that of the division part 122 to calculate an approximate value of b(x-Mx), and My is added by an addition part 124, and thus an estimated value corresponding to the signal 2 is obtained. This invention is applicable to linear regression arithmetic units or digital video cameras. COPYRIGHT: (C)2005,JPO&NCIPI
|