Anti-aliasing filtering of 2D images for multi-view auto-stereoscopic displays

Abstract

In this paper, we address the problem of anti-aliasing filtering of images to be displayed on auto-stereoscopic displays. Auto-stereoscopic displays are constructed to create 3D visual effect by no special glasses but utilizing extra optical layer to cast different images to different directions. The topology of such layer is a compromise between the number of different views generated and the spatial resolution per view being fraction of the full 2D spatial resolution. Usually, the compromise is achieved by slanted and non-rectangular sub-sampling grids causing however corresponding aliasing artefacts. These artefacts are especially visible and annoying when 2D imagery, such as graphics and text, is to be displayed on auto-stereoscopic displays. In our work, we design efficient anti-aliasing filters to mitigate this effect. Two classes of filters are studied for a 3D display case. The first class is the class of non-separable filters, which takes into account the non-rectangular topology of the particular sub-sampling grid and the effect of inter-view crosstalk, and aims at suppressing the respective aliasing replicas appearing in non-rectangular positions on the 2D Fourier plane. The second class is the class of efficient separable 2D filters based on 1D anti-aliasing filter design. We demonstrate that the latter class results in subjectively higher quality images. Studying this particular case further, we design filters for different types of imagery, distinguishing between text and graphics and also between ‘smooth’ and ‘sharp’ target anti-aliased images. As it is difficult to quantify the results by objective measures, we illustrate them by visual examples. Subjective inspections have also confirmed the feasibility of our approach.