Conventional Genetic Algorithms (CGAs) have the extensive application in image registration, but always have the problems of low accuracy and slow speed of convergence. By analyzing the mechanism of CGAs used in image registration, a notion called “retardation area” in GA search process was defined. Based on this point, we proposed a strategy of 8 Neighborhoods Search (8NS) in GA retardation area. This strategy forced the best individual to the direction of much higher fitness in GA retardation, which made the result of the image registration get higher accuracy. At the same time, the quality of gene was optimized. Hence, the quantity of similarity detection can be reduced and the speed of image registration can be speeded up. Compared with the exhaustive algorithm and CGAs, experimental results demonstrate that the 8NS-GA is faster and can achieve higher registration accuracy.