Filtering in a High Dynamic Range (HDR) Context

Many image processing steps involve filters or filter-like constructs (such as wavelets). Filters are used for displacement, such as flowfield displacement or block displacement within codecs. Filters are used for re-sizing, and for detail-band-splitting in resolution hierarchies. Wavelets also are applied in a manner similar to filters, and can be used for detail-band-splitting in some constructions. The Discrete Cosine Transform similarly represents image spatial frequencies in an array of coefficients for such frequencies. — When applying filters to pixels, theoretical underpinnings of filter theory often require that the pixels be represented in linear light (gamma 1.0). When extending filter and wavelet techniques to High Dynamic Range (HDR), the general assumptions about acceptability of filtering errors in many common filter uses must be revised to consider very large increases in brightness value differences. Filter errors, especially from negative filter lobes, are often greatly magnified within the intended viewing range. — This paper explores filtering practices that improve filter, wavelet, DCT, and other common filtering elements when applied to HDR. Further, we examine the precision necessary to distribute HDR, and we consider quality metrics that might be appropriate for HDR systems.

Published

2013-10

Content type

Original Research

Keywords

Digital Filters, HDR, Video Filtering

DOI

10.5594/M001509

ISBN

978-1-61482-953-9