Neural Network SDR to HDR Dynamic Range Conversion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing devices struggle to accurately convert input images from a standard dynamic range (SDR) to high dynamic range (HDR) due to insufficient statistical analysis, which fails to adapt to the specific features of the input image.
Innovation Solution
An image processing device utilizing a neural network-based mapping unit that transforms input luminance values into output luminance values, with a pre-processing module for statistical representation and an adaptive module for exponentiation, allowing for accurate feature mapping and adaptation to different image categories.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If analytical processing is used to determine tone expansion parameters, then the processing is simple and fast, but the conversion accuracy and adaptation to input image features are insufficient
Solution Approach 1:
The patent replaces traditional analytical processing methods with a neural network-based system. The neural network learns optimal tone expansion parameters through training on image data, substituting manual mathematical analysis with an automated machine learning approach that achieves higher accuracy while adapting to different image characteristics.
Solution Approach 2:
The patent changes the approach from fixed analytical formulas to dynamic parameter learning. The neural network adjusts conversion parameters based on the statistical properties of the input image, allowing the system to adapt parameters like gamma values and tone mapping curves according to the specific characteristics of each image rather than using universal fixed formulas.
2Adaptability or versatility
If fixed conversion parameters are used, then the processing is efficient, but the adaptability to different image categories and features is poor
Solution Approach 1:
The patent implements dynamic parameter adjustment through the neural network. Instead of using fixed conversion parameters, the system continuously adapts the tone expansion parameters based on the statistical representation of the input image, making the conversion process dynamic and responsive to different image characteristics while maintaining processing efficiency through the trained network.
Solution Approach 2:
The patent performs preliminary training of the neural network offline using representative image data. This preliminary action allows the network to learn optimal conversion strategies in advance, so that during actual processing, the system can quickly apply pre-learned parameters without extensive real-time computation, thus maintaining efficiency while achieving high adaptability.
3Measurement precision
If statistical analysis is performed on input image, then some image characteristics can be identified, but the features cannot be accurately transcribed for conversion
Solution Approach 1:
The patent replaces traditional statistical analysis methods with neural network-based feature extraction. The neural network processes the statistical representation of the input image and accurately transcribes the essential features into conversion parameters, preserving more image characteristic information than conventional statistical methods while reducing information loss in the conversion process.
Data Source
AI summary
An image processing device for converting an input image having a first dynamic range into an output image having a second dynamic range distinct from the first dynamic range includes a mapping unit for transforming, using a conversion parameter, an input luminance value associated with a pixel of the input image into an output luminance value associated with the corresponding pixel in the output image. The mapping unit includes a processing module based on a neural network, the neural network being configured to receive as an input a statistical representation depending on a plurality of input luminance values respectively associated with a plurality of pixels of the input image, the neural network being configured to provide as an output the conversion parameter. A method for converting an input image into an output image is also provided.