HDR Image Generation With Region-Specific Motion Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing imaging systems struggle to capture high dynamic range (HDR) images effectively due to motion-related noise and misalignment issues, particularly in self-portraits and forward-facing images, where rotational and translational displacements during multiple exposures lead to ghosting and misalignment of foreground and background objects.
Innovation Solution
A method for generating HDR images using multi-domain motion correction, involving the use of gyroscope sensors to detect rotational movements and translational displacements, and applying transformation matrices to align and combine images with different exposures, specifically focusing on foreground and background regions to correct for rotational and spatial misalignments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Illumination intensity
If multiple exposures with different exposure times are used to capture HDR images, then the dynamic range and image quality are improved, but motion-related noise and misalignment issues occur due to rotational and translational displacements during capture
Solution Approach 1:
The image is divided into multiple regions (foreground objects, background objects, and border regions) with different transformation requirements. Different transformation matrices are applied to different regions based on their depth and motion characteristics, allowing precise alignment while preserving HDR quality.
Solution Approach 2:
Different transformation matrices are applied to different regions of the image. Foreground objects use one transformation matrix, background objects use another, and border regions use interpolated transformations. This local differentiation resolves the alignment precision issue while maintaining overall HDR image quality.
2Manufacturing precision
If transformation matrices are applied to correct motion in HDR images, then alignment precision is improved, but computational complexity and processing time increase
Solution Approach 1:
The transformation process is segmented into distinct regions (foreground, background, border) with different computational requirements. This allows the system to apply complex transformations only where necessary while using simpler transformations elsewhere, reducing overall computational complexity.
Solution Approach 2:
Instead of applying a single complex transformation to the entire image, the system applies partial transformations to specific regions. The border regions use interpolated transformations that are computationally lighter than full transformations, reducing processing complexity while maintaining alignment precision.
3Manufacturing precision
If region-based transformations are applied to foreground and background objects, then alignment precision and noise reduction are improved, but the complexity of detecting and measuring motion increases
Solution Approach 1:
The image is segmented into foreground and background regions using depth information from the neural network. This segmentation simplifies motion detection by focusing on specific regions rather than the entire image, reducing the difficulty of detecting and measuring motion while improving alignment precision.
Solution Approach 2:
A neural network serves as an intermediary to predict depth maps and identify foreground objects. This intermediary simplifies the motion detection process by providing pre-processed depth information that guides the region-based transformation, reducing the overall complexity of motion detection and measurement.
4Manufacturing precision
If deep learning neural networks are used to predict depth maps and identify foreground objects, then manufacturing precision of alignment is improved, but use of energy and computational resources increases
Solution Approach 1:
The neural network predicts depth maps and identifies foreground objects in advance, before the transformation process. This preliminary action provides ready-to-use depth information that simplifies subsequent transformation operations, reducing the overall computational energy required while maintaining high alignment precision.
Solution Approach 2:
The neural network segments the image into foreground and background regions based on depth prediction. This segmentation allows the system to apply transformations only to relevant regions, reducing unnecessary computational energy while improving alignment precision through region-specific processing.
Data Source
AI summary
Disclosed are systems, apparatuses, processes, and computer-readable media to capture images with subjects at different depths of fields. A method of processing image data includes obtaining a first image captured using an image sensor, the first image being associated with a first exposure: obtaining a second image captured using the image sensor, the second image being associated with a second exposure that is longer than the first exposure: modifying a first region of the first image based on a first transformation and a second region of the first image based on a second transformation to generate a modified first image; and generating a combined image at least in part by combining the modified first image and the second image.