Joint Dictionary Generation for Interlace HDR Imaging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Interlace-based HDR imaging faces challenges with spatial resolution reduction and noise issues, particularly in dark areas, due to the need for different exposure settings in odd and even macro rows, which can lead to the 'ghost effect' and severe noise problems.
Innovation Solution
The implementation of a joint dictionary generation method using joint dictionary learning to generate high-quality and low-quality dictionaries that share a sparse code, allowing for the removal of interlace artifacts and noise from high-exposure and low-exposure RGB frames, thereby enhancing the quality of HDR images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If interlace-based HDR imaging is used to capture images at different exposures simultaneously, then ghost effect is reduced, but spatial resolution is reduced to half
Solution Approach 1:
The patent combines high-resolution and low-resolution image data through joint dictionary learning to reconstruct images that maintain both the ghost effect reduction benefits of interlaced capture and the spatial resolution of full-resolution imaging. The method merges complementary information from different exposure images to achieve high-quality HDR reconstruction.
2Reliability
If gain is adjusted for each macro row to change exposure, then HDR imaging is enabled, but severe noise is caused in high gain macro rows
Solution Approach 1:
The patent changes the parameter representation from direct pixel values to sparse code coefficients in a learned dictionary space. This transformation allows noise suppression by separating signal from noise in the sparse domain, enabling effective denoising while preserving HDR imaging capability across different exposure levels.
Solution Approach 2:
The patent introduces joint dictionaries as an intermediary representation between the raw interlaced image data and the final HDR output. These dictionaries serve as a mediator that enables simultaneous denoising and resolution enhancement by providing a structured sparse representation that captures underlying image structures while filtering noise.
3Manufacturing precision
If joint dictionary learning is performed to remove interlace artifacts and noise, then image quality is improved, but processing complexity increases
Solution Approach 1:
The patent performs preliminary dictionary learning on training data to pre-compute the joint dictionaries that will be used for reconstruction. This preliminary action separates the complex learning phase from the actual image processing phase, allowing the reconstruction to use the pre-learned dictionaries for more efficient processing while still achieving high image quality.
Data Source
AI summary
An HDR imaging apparatus include: a deinterlacing circuit configured to generate a high-exposure Bayer frame and low-exposure Bayer frame by deinterlacing an interlace Bayer raw frame in which a high-exposure region and low-exposure region are interlaced; a demosaicing circuit configured to convert the high-exposure Bayer frame and the low-exposure Bayer frame into a high-exposure RGB frame and low-exposure RGB frame, respectively; a reconstructing circuit configured to remove interlace artifacts and noise from the high-exposure RGB frame and the low-exposure RGB frame, using joint dictionaries generated through joint dictionary learning on training data sets each containing a plurality of bases, but the corresponding bases have different values; and an HDR generation circuit configured to generate an HDR image frame by combining the high-exposure RGB frame and the low-exposure RGB frame from which interlace artifacts and noise were removed by the reconstructing circuit.