| Abstract: | A simple attention-based model is proposed for efficient transmission of visual information using multiresolution structures. Images are sampled nonuniformly in space and time, such that sampling is dense at the focus of attention and sparse in the periphery (retinal-like). Assuming that the focus of attention usually corresponds to eye position while scanning an image, image features which are “eye catching” (such as sharp edges, motion, and high flicker rate) are used to drive the dense center of sampling. The transmitted image is reconstructed by combining each new sample with previous samplings to give progressive transmission. The selection of sampling points depends only on previously transmitted information, and only sampled values without their location need to be transmitted. |
