Content-Adaptive Perceptual Quantization for HDR Image Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current technologies face challenges in efficiently processing high-dynamic-range images, particularly in terms of perceptual quantization, which affects image quality and compatibility across different display devices with varying dynamic ranges.
Innovation Solution
The implementation of Content-Adaptive Perceptually Quantized (CAQ) image processing techniques, including forward and backward reshaping functions, dynamic range non-linear scaling, and noise masking, to optimize bit depth allocation and image merging/blending, ensuring efficient encoding and decoding across a wide range of dynamic ranges.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If perceptual quantization is applied to HDR images, then image quality is improved, but compatibility across different display devices deteriorates
Solution Approach 1:
The patent segments the dynamic range into multiple bands (e.g., 5 bands) and applies different quantization strategies to each band. This allows fine-grained control over quantization precision in different luminance regions, improving overall image quality while maintaining adaptability to various display devices through selective application of quantization parameters.
Solution Approach 2:
The patent employs dynamic range non-linear scaling that adapts to different display device characteristics. By using content-adaptive techniques that adjust quantization parameters based on the actual content and target display properties, the system maintains high image quality across diverse display devices with varying dynamic ranges.
2Measurement precision
If high-precision floating-point formats are used for HDR images, then dynamic range representation is improved, but data transmission and processing efficiency deteriorates
Solution Approach 1:
The patent transforms HDR image data from high-precision floating-point format to optimized integer representations using perceptual quantization. By changing the numerical representation parameters based on human visual system characteristics and display device capabilities, the system achieves efficient data transmission and processing while preserving perceptual image quality.
Solution Approach 2:
The patent extracts and removes redundant high-precision information that is not perceptually significant. Through perceptual quantization, the system identifies and eliminates excessive precision in regions where the human visual system cannot distinguish differences, thereby reducing data size and improving transmission efficiency without noticeable quality loss.
3Device complexity
If standard quantization methods are used for HDR images, then processing simplicity is maintained, but noise artifacts increase
Solution Approach 1:
The patent applies local quality enhancement through content-adaptive quantization that adjusts parameters based on local image characteristics such as luminance level, noise content, and texture complexity. This allows the system to maintain processing simplicity while reducing noise artifacts by applying different quantization strategies to different regions of the image.
Solution Approach 2:
The patent converts potential noise artifacts into beneficial effects by using perceptual quantization that aligns with human visual system characteristics. By quantizing in a way that matches perceptual thresholds, the system transforms what would be noticeable noise into imperceptible quantization steps, effectively converting a harmful artifact into an imperceptible feature.
Data Source
Figure 1A~1B
Figure 2
Figure 3A
AI summary
An image processing device receives one or more forward reshaped images that are generated by an image forward reshaping device from one or more wide dynamic range images based on a forward reshaping function. The forward reshaping function relates to a backward reshaping function. The image processing device performs one or more image transform operations on the one or more forward reshaped images to generate one or more processed forward reshaped images without performing backward reshaping operations on the one or more reshaped images or the one or more processed forward reshaped images based on the backward reshaping function. The one or more processed forward reshaped images are sent to a second image processing device.