Graph Signal Processing for HDR Image Ghosting Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Digital cameras face challenges in capturing high dynamic range images due to limited sensor dynamic range, leading to artifacts like ghosting from inappropriate image registration and object movement, especially in hand-held cameras.
Innovation Solution
A method and apparatus for obtaining a high dynamic range image by processing multiple low dynamic range images, where weighting values are determined based on pixel similarity within blocks to assign new pixel values, using graph signal processing techniques to align and combine images effectively.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Illumination intensity
If multiple LDR images are captured and combined using traditional HDR techniques, then the dynamic range of the final image is improved, but ghosting artifacts occur due to camera movement and object movement during capture
Solution Approach 1:
The image is divided into multiple blocks, and each block is processed independently to determine weighting values based on local pixel similarity. This segmentation allows the algorithm to handle different regions with different characteristics, reducing the impact of camera movement and object movement on the overall HDR quality.
Solution Approach 2:
The patent applies local quality by computing weighting values based on pixel similarity within local blocks rather than globally. This allows the HDR combination to adapt to local variations in the scene, preserving details in static regions while handling moving objects and camera motion artifacts more effectively.
2Ease of manufacture
If traditional image registration is used to align multiple LDR images, then the alignment process is simple, but ghosting artifacts occur due to inappropriate registration
Solution Approach 1:
The patent replaces traditional mechanical image registration with a graph signal processing approach. Instead of relying on complex registration algorithms that may fail with hand-held cameras, the method uses graph-based pixel similarity measures to implicitly handle alignment, substituting a more robust but computationally different approach.
3Object-affected harmful factors
If graph signal processing is used to process pixel blocks, then ghosting artifacts are reduced and image quality is improved, but the computational complexity increases
Solution Approach 1:
By segmenting the image into blocks and processing each block independently with graph signal processing, the patent reduces the overall computational complexity compared to applying graph signal processing to the entire image at once. This segmentation makes the complex algorithm more tractable while preserving its artifact-reduction benefits.
Data Source
AI summary
A method of obtaining one or more HDR images representative of a scene is described. To reach that aim, the method includes obtaining several LDR images representative of the scene. The method further includes identifying one or more first pixels having a pixel value that is greater than a first determined value (i.e. corresponding to underexposed pixels) or having a pixel value that is less than a second determined value (i.e. corresponding to overexposed pixels). The method further includes determining one second pixel, in one or more other LDR images, that corresponds to the first pixel. In a block of pixels centered on the second pixel, weighting values are determined that are representative of similarity between the second pixel and the other pixels of the block. A HDR image is finally obtained by determining and assigning a new pixel value to the first pixel. The new value is calculated based on weighting values and pixel values associated with pixels of a block of pixels centered on the first pixel.