This paper addresses the problem of stereo image pair compression in remote sensing applications. Three fundamental techniques including Disparity Compensation （DC）

Radiation Compensation （ RC） and the compression of residual image are discussed. The main issue of stereo image compression is the estimation of disparity between the left frame and the right frame

a problem similar to the estimation of motion vectors in video coding. Since the disparity consisting in remote sensing image pairs are caused by variation in elevation rather than motion of the some objects

it is generally more universal and complex than the motion vectors. However

due to the geometrical constraint between the two frames

the vertical disparity vectors of remote sensing stereo image pairs can always be eliminated by a certain relative-orienting method. Besides stereo disparity

the average grayscale difference between corresponding blocks is also a big obstacle for creating accurate predictive image

for the two frames of a remote sensing image pair usually have considerable luminance disparity due to the changing in photograph angle and incidence angle. To fully exploit the redundancy between the two frames

a Stereo Compensation （SC） technique combining adaptive DC with RC is presented in this paper. It shows good performance in predicting the right frame from the left frame. A high-performance integer wavelet image coder is utilized for the compression of residual image. Experiments show that the SC-based compression algorithm for remote sensing stereo image pair is about 5% and 30%-45% more efficient than JPEG2000 in lossless compression and lossy compression respectively.