Inverse Tone Mapping CNN for HDR Detail Restoration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional inverse tone mapping methods are unable to fully utilize the capabilities of HDR TVs due to weaknesses in generating full contrast and details, or noise amplification, when applied to HDR displays with a maximum brightness of 1,000 cd/m2.
Innovation Solution
An effective convolutional neural network (CNN) architecture, ITM-CNN, is proposed for up-converting single 8-bit/pixel gamma-corrected LDR images to 10-bit/pixel HDR images using perceptual quantization transfer function in the BT.2020 color container, enabling direct viewing on commercial HDR TVs without additional information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If conventional ITM methods are applied to HDR TV displays, then the method can process LDR images, but it fails to generate full contrast and details and amplifies noise
Solution Approach 1:
The patent changes the fundamental parameters of the ITM process by operating in the logarithmic domain instead of linear domain, and by using a learned non-monotonic mapping function instead of conventional monotonic tone mapping curves. This allows the method to preserve noise while enhancing contrast and details simultaneously.
Solution Approach 2:
The patent inverts the conventional tone mapping approach by using inverse tone mapping to go from LDR to HDR domain, rather than the traditional tone mapping from HDR to LDR. This inversion allows direct generation of HDR images with full contrast and detail preservation.
2Manufacturing precision
If conventional ITM methods are applied to HDR TV displays, then the method can process LDR images, but it fails to generate full contrast and details
Solution Approach 1:
The patent changes the mathematical parameters by using logarithmic domain processing and non-monotonic mapping functions, enabling simultaneous enhancement of contrast and preservation of details that conventional monotonic tone mapping cannot achieve.
Solution Approach 2:
By inverting the conventional tone mapping direction and using inverse tone mapping in the logarithmic domain, the patent achieves direct synthesis of HDR images with full contrast and detail, rather than approximating them through traditional methods.
3Adaptability or versatility
If conventional ITM methods are applied to HDR TV displays, then the method can process LDR images, but it cannot fully utilize the available HDR capacity
Solution Approach 1:
The patent changes the operational parameters to match HDR display capabilities by using 10-bit depth, logarithmic domain processing, and non-monotonic mapping functions, enabling full utilization of HDR capacity for contrast and detail representation.
Solution Approach 2:
The inverse tone mapping approach directly synthesizes HDR images optimized for display on HDR TVs, rather than creating intermediate representations that require further processing, thus fully utilizing available HDR capacity.
Data Source
AI summary
In this invention, we propose a convolutional neural network (CNN) based architecture designed for the ITM to HDR consumer displays, called ITM-CNN, and its training strategy for enhancing the performance based on image decomposition using the guided filter. We demonstrate the benefits of decomposing the image by experimenting with various architectures and also compare the performance for different training strategies. To the best of our knowledge, this invention first presents the ITM problem using CNNs for HDR consumer displays, where the network is trained to restore lost details and local contrast. Our ITM-CNN can readily up-convert LDR images for direct viewing on an HDR consumer medium, and is a very powerful means to solve the lack of HDR video contents with legacy LDR videos.