Beyond blur

To peripheral vision, a pair of physically different images can look the same. Such pairs are metamers relative to each other, just as physically-different spectra light perceived same color. We propose real-time method compute such ventral for foveated rendering where, in particular near-eye displays, largest part framebuffer maps periphery. This improves quality over state-of-the-art foveation methods which blur Work Vision Science has established how stimuli if their statistics similar. Existing methods, however, require costly optimization process find metamers. this end, we novel type particularly well-suited practical rendering: smooth moments steerable filter responses. These be extracted from time constant number pixels and parallel all using GPU. Further, show that they compressed effectively transmitted at low bandwidth. Finally, computing realizations those again performed parallel. enables new level applications remote rendering, level-of-detail Monte-Carlo denoising. In user study, finally human task performance increases artifacts less suspicious, when our compared common blurring.