Matching the Human Visual System, Balancing Bit Depth, High Dynamic Range and Coding Efficiency

The human visual system sensitivity is not linear: it requires much more detail in the low values. This phenomenon has been generally modelled by a logarithmic relation between magnitude and sensation. — This aspect of the visual sensation has always been implemented in various ways in high quality imaging systems, from the 2.2 gamma in legacy video to logarithmic coding for film scanned content and 2.6 coding for Digital Cinema. — As a result, we have multiple domains with complicated exchange, specific computations with ad hoc parameters, rounding errors and degradation arising from multiple coding decoding concatenations. — It is possible to avoid these complications and have a unified system by using “Unsigned floating point” which is a simple extension to floating point “half” format. The additional benefits are a very high dynamic range, a built-in logarithmic law and an efficient use of codes.

Published

2011-10

Content type

Original Research

Keywords

Human Visual System, dynamic range, coding efficiency, perception, floating point, pixel format

DOI

10.5594/M001096

ISBN

978-1-61482-940-9