Space Variant Imaging
Center for Perceptual Systems
The University of Texas at Austin
Foveated MPEG refers to a method of image compression where each image in an image sequence is first passed through a foveated lowpass pyramid before being encoded by an MPEG encoder. On the decoder side, the images are first MPEG decoded, and then reconstructed with the foveated lowpass pyramid. The effect of applying the foveated lowpass image filter is to reduce the total number of pixels processed by the MPEG encoder for a given sequence. Thus, foveated MPEG can be used to obtain higher compression ratios, and hence reduce the data rate requirements.
The following chart compares data rates and compression ratios of a foveated and unfoveated MPEG image sequence.
Consider the typical video communication environment where transmission bit rate is held constant. In this situation, foveated MPEG can be used to enhance the image quality in areas of interest. This enhancement occurs because a foveated MPEG sequence contains fewer pixels, and hence does not require as much quantization as a standard MPEG sequence to achieve the same bit rate. (Alternatively, foveated MPEG can be used to increase the frame rate rather than enhance the image quality.)
The following images demonstrate the concept of enhancing image quality in areas of interest at constant bit rates. The first image in each pair of frames was taken from a foveated MPEG image sequence. The second image in each pair of frames was taken from a standard MPEG image sequence.
Both sequences were compressed at approximately equal data rates: about 80 Kbps at 30 frames per second. However, in the foveated sequences, the total amount of pixels to compress was less than in the unfoveated sequences. Therefore, in order to achieve equal bit rates, the foveated sequences required less quantization. This can be confirmed by visually comparing the area of interest in corresponding frames: The woman in frame 1 of the foveated MPEG sequence (first image) is of higher visual quality than in frame 1 of the standard MPEG sequence (second image).
Frame 1: High resolution is concentrated on the woman in the first image. Resolution is constant in the second image.
Frame 70: High resolution is concentrated on the man in the first image. Resolution is constant in the second image.
Frame 95: High resolution is concentrated on the dancer in the first image. Resolution is constant in the second image.
Copyright (C) 2002-2019, Center for Perceptual Systems|