Efficient Dataflow Modeling of Peripheral Encoding in the Human Visual System

Computer graphics seeks to deliver compelling images, generated within a computing budget, targeted at specific display device, and ultimately viewed by an individual user. The foveated nature of human vision offers opportunity efficiently allocate computation compression appropriate areas the viewer’s visual field, particular importance with rise high-resolution wide field-of-view devices. However, while variations in acuity contrast sensitivity across field view have been well-studied modeled, more consequential variation concerns peripheral vision’s degradation face clutter, known as crowding. Understanding crowding has greatly advanced recent years, terms both phenomenology modeling. Accurately leveraging this knowledge is critical for many applications, covers majority pixels image. We advance computational models aimed toward their eventual use computer graphics. In particular, researchers recently developed high-performing crowding, “pooling” models, which predict range phenomena but are computationally inefficient. reformulate problem dataflow computation, enables faster processing operating on larger images. Further, we account explicit encoding “end stopped” features image, was missing from previous methods. evaluate our model context perception textures periphery, including novel texture dataset updated textural descriptors. Our improved framework may simplify development testing sophisticated, complete robust realistic settings relevant