Abstract: As a linear blending method, cage-based deformation is widely used in various applications of image and geometry processing. In most cases especially in the interactive mode, deformation based on embedded cages does not work well as some of the coefficients are not continual and make the deformation discontinuous, which means existing "spring up" phenomenon. However, it's common for us to deform the ROI(Region of Interest) while keeping local part untouched or with only small adjustments. In this paper, we design a scheme to solve the above problem. A multicage can be generated manually or automatically, and the image deformation can be adjusted intelligently according to the local cage shape to preserve important details. On the other hand, we don't need to care about the pixels' position relative to the multicage. All the pixels go through the same process, and this will save a lot of time. We also design a packing method for cage coordinates to pack all the necessary coefficents into one texture. A vertex shader can be used to accelerate the deformation process, leading to realtime deformation even for large images.