Abstract: This paper presents a bi-graph image decomposition based on quincunx sampling lifting scheme, which decomposes an original image to a lower-resolution one and a different one between the original image and the lower-resolution image. The proposed scheme is of low-complexity and need not allocate additional memory.This paper introduces some simple examples, such as linear mean-lifting, nonlinear max-lifting and min-lifting. Mean-lifting is good at erasing redundant data, and max-lifting or min-lifting can effectively preserve important geometric information.