HDR Video Compression Using Perceptual Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current lossy compression techniques for high dynamic range (HDR) images face challenges such as limited bit depth per color channel, imprecision in RGBE encoding, and inefficiency in compressing negative numbers, particularly when using MPEG or JPEG compression, which trade off precision against dynamic range and are slow or complex.
Innovation Solution
The method involves mapping floating-point image data from a light linear space to a perceptually linear space using a piecewise power and logarithmic function, followed by a discrete cosine transform (DCT), and then quantizing and losslessly compressing the DCT frequency coefficients, allowing for direct compression of floating-point values without converting to integer formats.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If HDR image data is converted from floating point format to limited-range integer format for compression, then compression efficiency is improved, but precision is lost and dynamic range is reduced
Solution Approach 1:
The patent applies parameter changes by using perceptual mapping functions (such as piecewise power functions or logarithmic functions) to transform HDR image data from linear light space to perceptually uniform space. This transformation changes the numerical parameters of the image data while preserving visual information, allowing the data to be compressed more efficiently after the mapping without losing perceptually relevant precision
Solution Approach 2:
The patent segments the HDR image data processing into distinct stages: first applying perceptual mapping to transform the data, then performing compression on the transformed data. This segmentation allows each stage to be optimized independently - the mapping stage preserves precision while the compression stage improves efficiency
2Productivity
If perceptual mapping is applied to HDR image data before compression, then compression performance is improved, but computational complexity increases
Solution Approach 1:
The patent uses perceptual mapping functions that transform image data parameters in a computationally efficient manner. These mapping functions (such as piecewise power functions with predetermined coefficients or logarithmic functions) are designed to provide good perceptual results while maintaining reasonable computational complexity, avoiding overly complex transformations
3Ease of manufacture
If traditional compression techniques like MPEG or JPEG are used on HDR data, then implementation is simpler, but they trade off precision against dynamic range and are slow to run
Solution Approach 1:
The patent applies preliminary action by performing perceptual mapping on the HDR image data before applying compression algorithms. This preprocessing step transforms the data into a format that is more suitable for compression, improving the efficiency of subsequent compression operations while maintaining implementation simplicity through the use of well-established compression techniques
Data Source
AI summary
A high dynamic range (HDR) video can be compressed using inter-frame or intra-frame compression. The compression results in multiple encoded macroblocks and associated residual images. The residual images can be lossy compressed by mapping floating point image data values of the residual image from a light linear space to a perceptually linear space using a first non-linear mapping function. A value in the light linear space is proportionate to the power of a light source, and a value in the perceptually linear space is proportionate to a visual perceived brightness. Floating point discrete cosine transform (DCT) frequency coefficients of the floating point image data values of the residual image are determined. The floating point DCT frequency coefficients are quantized. The quantized floating point DCT frequency coefficients are compressed. The compression of the quantized floating point DCT frequency coefficients is lossless. The compressed quantized floating point DCT frequency coefficients are stored.