Predictive Coding Dynamic Range Conversion for High Dynamic Range Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Standard image coding methods are inefficient for handling high dynamic range images greater than 14 bits, as they require conversion of pixel values to a lower dynamic range, leading to loss of temporal coherency and reduced prediction quality due to differing quantization parameters across groups of pictures.
Innovation Solution
A predictive coding/decoding method that applies a dynamic conversion transform to each group of pictures to align pixel values with a target dynamic, followed by inter-image prediction and inverse transformation, ensuring consistent quantization parameters across reference and current images, and using tone mapping to minimize quantization errors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If standard image coding methods are used for high dynamic range images, then compression can be achieved, but temporal coherency is lost and prediction quality deteriorates due to differing quantization parameters across groups of pictures
Solution Approach 1:
The patent applies a configurable transform to convert pixel values to a target dynamic range before coding, and crucially, applies the same transform to reference images from previous groups of pictures. This preliminary action ensures that both current and reference images have consistent quantization parameters, maintaining temporal coherency while enabling effective compression.
Solution Approach 2:
The patent changes the dynamic range parameter of pixel values by applying a configurable transform that maps values from the source dynamic range to a target dynamic range. This parameter change ensures uniform quantization across all images in the sequence, resolving the contradiction between maintaining temporal coherency and achieving compression efficiency.
2Adaptability or versatility
If pixel values are converted to lower dynamic range for standard coding methods, then compatibility with standard coders is achieved, but quantization errors increase and prediction quality reduces
Solution Approach 1:
The patent applies a configurable transform that changes the dynamic range parameter of pixel values to match the target dynamic range of standard coding methods. This ensures compatibility while minimizing quantization errors through optimized transform parameters that preserve prediction quality.
Solution Approach 2:
The patent uses reference images from previous groups of pictures that have been processed through the same transform to generate predictions for current images. This feedback mechanism allows the system to maintain high prediction quality by leveraging consistent quantization parameters across temporal references.
3Adaptability or versatility
If different quantization parameters are used for different groups of pictures, then adaptation to varying image characteristics is improved, but motion field uniformity deteriorates and motion vector coding cost increases
Solution Approach 1:
The patent applies a configurable transform with parameters determined by at least one image of the group of pictures, allowing adaptation to varying image characteristics. By ensuring the same transform is applied to both current and reference images, the patent maintains uniform motion fields, reducing motion vector coding complexity.
Data Source
Figure 1~2
Figure 3~4
Figure 5
AI summary
The invention relates to a predictive coding/decoding method/device for a group of pictures belonging to a sequence of images, using at least one reference image belonging to a group of pictures other than the group of pictures to code, and converting the values of the pixels of the images of the group of pictures from a configurable transform such that these values are expressed in a target dynamic. The method is characterised in that it converts the values of the pixels of each reference image from the transform thus configured identically to the one used to convert the values of the pixels of the images of the group of pictures to code.