A new type of scene adaptive coder has been developed. The described coder involves a quadtree mean decomposition of the motion compensated frame-to-frame difference signal followed by a scalar quantization of the local means. As a fundamental property, the new coding algorithm treats the displacement estimation problem and the quadtree construction problem as a unit. The displacement vector and the related quadtree are jointly optimized in order to minimize the direct frame-to-frame update information rate (in bits) which turns up as a new and more adequate cost function in displacement estimation. This way the highest possible data compression ratio at a given quality threshold is always guaranteed. Excellent results have been obtained for coding of color image sequences at a rate of 64 kbits/s. The presented quadtree concept attracts with a much lower computational complexity when compared to the conventional motion compensated transform coder while achieving a subjective image quality that is as good or better than the traditional transform-based counterpart. © 1990 IEEE