Presented at the 1999 Texas Instruments Digital Signal Processing Systems Fest

Efficient Implementation of Foveation Filtering

Sanghoon Lee, Alan C. Bovik, and Brian L. Evans

Laboratory for Image and Video Engineering, Engineering Science Building, The University of Texas at Austin, Austin, TX 78712-1084
bovik@ece.utexas.edu - bevans@ece.utexas.edu

Paper - Slides

Abstract

In the human eye, the maximum detectable frequency is a function of the photoreceptor density and eccentricity (visual angle). The maximum detectable frequency is obtained by using an exponential model of the spatial sampling grid on the fovea in the retina and choosing the model parameters such that the human eye cannot discriminate the foveated image from the original when assuming a given fixation point in the image being observed. It can also be used as the cutoff frequency for obtaining a foveation-filtered image, which is a nonuniform resolution image with spatial resolution at its peak near the fixation/foveation point and falling off away from this point. Foveation filtering can be implemented using a bank of lowpass filters. In order to reduce the number of operations, we use separable, even-symmetric or circularly symmetric lowpass filters. The number of taps in each filter is adaptively changed according to the magnitude of the cutoff frequency. The overall design is amenable to incorporation into implementations of foveated H.263 compression algorithms on digital signal processors.


Last Updated 05/18/09.