Floating-Point Image Compression with Quantized Luma-Chroma Streams
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image compression techniques face challenges in efficiently compressing floating-point format images, particularly those with a wide-gamut color space, as they are not optimized for the unique characteristics of floating-point values, leading to suboptimal compression ratios and increased computational overhead.
Innovation Solution
The proposed solution involves pre-processing floating-point format multiple-channel images by quantizing pixels, applying invertible color-space transformations, separating data streams, and using predictive functions to enhance compression ratios, allowing for efficient compression and decompression in hardware and software optimized for wide-gamut color spaces.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If higher compression ratios are used to increase storage space, then storage capacity is improved, but compression and decompression computational overhead increases
Solution Approach 1:
The patent applies preliminary quantization to floating-point image data before compression, converting values to a fixed-point representation. This pre-processing step simplifies the subsequent compression process by eliminating the need to handle variable-precision floating-point numbers during compression, thereby achieving higher compression ratios without proportionally increasing computational complexity.
Solution Approach 2:
The patent transforms the data representation parameter from floating-point to fixed-point format through quantization. This parameter change enables more efficient compression by reducing the complexity of the numerical representation, allowing standard compression algorithms to achieve better ratios without excessive computational overhead.
2Reliability
If lossless compression is applied to floating-point images, then image quality is preserved, but compression efficiency decreases
Solution Approach 1:
The patent changes the numerical representation parameter from floating-point to fixed-point through quantization, which enables more efficient compression algorithms to be applied while maintaining lossless compression capabilities. The fixed-point representation has predictable bit patterns that compression algorithms can exploit more effectively.
Solution Approach 2:
The patent substitutes the complex floating-point arithmetic system with a simpler fixed-point system for the compression process. This substitution allows standard compression algorithms to operate more efficiently on the quantized data while preserving the ability to reconstruct the original image losslessly.
3Ease of manufacture
If standard compression algorithms are used on floating-point images, then implementation is simple, but compression ratios are suboptimal
Solution Approach 1:
The patent introduces a preliminary quantization step that transforms floating-point data into fixed-point format before applying standard compression algorithms. This pre-processing maintains implementation simplicity by still using standard algorithms, while significantly improving compression ratios through the optimized data representation.
Solution Approach 2:
The patent modifies the data parameter from floating-point to fixed-point representation, which creates more favorable conditions for standard compression algorithms to achieve better compression ratios. The fixed-point format provides more predictable statistical properties that compression algorithms can exploit.
Data Source
AI summary
Disclosed herein are techniques for pre-processing a multiple-channel image for compression. The multiple-channel image can be composed of a collection of pixels that are represented using a floating-point format (e.g., half-precision/16-bit) for display on devices optimized for wide-gamut color space. The techniques can include a first step of quantizing the pixels into a fixed range of values, and applying invertible color-space transformations to the sub-pixels of each pixel—which can include red, green, blue, and alpha sub-pixels—to produce transformed sub-pixels including luma and chroma values. Next, the luma sub-pixels are placed into a luma data stream, the first and second chroma values are placed into a chroma data stream, and the alpha sub-pixels are placed into an alpha data stream. Predictive functions are then applied to the luma and chroma data streams. Finally, the various streams are separated into buffers and compressed to produce a multiple-channel image.