We propose a general framework for computing invariant features from images.The proposed approach is based on a simple concept of basis expansion. It iswidely applicable to many popular basis representations, such as waveletsshort-time Fourier analysis, and splines. Exploiting formulations that useboth global and local information about shape and color, the new approach isneither strictly global nor local. It has the advantage of tolerating acertain degree of occlusion (unlike global analysis) and does not requireestimating high-order derivatives in computing invariants (unlike localanalysis), whence is more robust. Furthermore, it enables a quasi-localized,hierarchical shape analysis which is not possible with other known invarianttechniques. Unlike most current research on image invariants whichconcentrates on either geometry or illuminance invariants, the proposedframework is very general and produces invariants which are insensitive torigid motion, general affine transform, changes of parameterization and sceneillumination, and perspective transform.