Image pyramid is an important tool for processing and analyzing digital images. This study focuses on image pyramid

and finds out an approach to reduce the data volume of them and solve the problem of truncation in precision. Based on a same idea

we propose two kinds of image pyramid: the mean-like pyramid and the Gaussian-like pyramid. In them

father-pixels can be got by simple calculations of their son-pixels; there is no problem of truncation in such calculations

and no extra bits are needed to store decimals. Besides

as some son-pixels can be recovered from their father-pixels

they can be dropped when storing a pyramid. Thereby

the number of pixels need to be stored is equal to that of the original image only. Its shown that the proposed pyramid structures have good performance in general by theoretic analysis and experimental on precision

data volume and speed of creation.