A Simple Method for Foveated Imaging

Abstract

　Human visual system has a space-variant resolution nature. In the retinal receptive field, the resolution is not uniform but sampled finer in the central fovea and coarser in the peripheral. This variable resolution mapping function is born by the cerebral primary visual cortex V1. It has a clear visual field map of spatial information, and this spatial mapping structure is called Retinotopy. The forward mapping to visual cortex from retina is characterized with complex LPT （Log-Polar-Transform） by Schwartz. The retinal receptive field image is reconstructed by inverse projecttion LPT -1 from V1. This reconstruction process is called Foveation. Since the spatial information is concentrated in the center of the visual field, the foveation is applied to image compression, pattern recognition, robot vision, and/or computer vision. In the previous report, we introduced the basic characteristics of Foveation, its application to region extraction, and the visual display effect. The retinal receptive field image is suitable for material appearance expression with natural blurring due to peripheral vision. However, the complexity of the inverse transformation LPT -1 was a bottleneck. The following report proposed a fast Foveation method using a Gaussian blurring logarithmic ring mask. Though, generation and synthesis of multi-stage ring masks are still complicated, and further simplification is desired. This paper proposes a novel and simple Foveation method using positive and negative Gaussian blur masks without using the inverse transform LPT -1 of Schwartz theory and also reports the quantitative evaluation results on reproduction errors.