Wavelet HDR Imaging Without Optical Flow Artifacts
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing HDR imaging methods suffer from artifacts like ghosting, missing details, color degradation, and noise, and are computationally inefficient, making them unsuitable for deployment on wearable devices.
Innovation Solution
A method using discrete wavelet transform and inverse discrete wavelet transform combined with a U-Net architecture for HDR image generation, employing frequency-specific component processing and attention-guided feature merging to align and fuse LDR images without optical flow, reducing information loss and computational cost.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If optical flow is used to align LDR images, then alignment accuracy is improved, but computational cost increases and hardware compatibility decreases
Solution Approach 1:
The patent extracts and removes the optical flow alignment step from the HDR imaging pipeline. Instead of using optical flow to align LDR images, the invention directly processes images in the frequency domain using wavelet transform, eliminating the complex alignment computation while maintaining processing effectiveness through frequency-domain operations.
Solution Approach 2:
The patent substitutes the mechanical/optical alignment process (optical flow) with a mathematical transformation approach (wavelet transform in frequency domain). This replacement eliminates the need for complex computational alignment while achieving the desired image fusion through frequency-specific component processing.
2Manufacturing precision
If multiple LDR images with different exposures are merged, then HDR image quality is improved, but artifact generation increases
Solution Approach 1:
The patent segments the LDR images into frequency-specific components using wavelet transform, separating them into different frequency bands (low-frequency and high-frequency components). This segmentation allows independent processing of different frequency components, enabling precise control over how information from multiple exposures is combined, thereby reducing artifacts while maintaining HDR quality.
Solution Approach 2:
The patent applies different processing strategies to different frequency components of the images. Low-frequency components are handled with attention to overall structure and alignment, while high-frequency components are processed to preserve edges and details. This local quality approach ensures that each frequency band contributes optimally to the final HDR image, minimizing artifacts.
3Measurement precision
If deep neural networks are used for HDR imaging, then image processing capability is improved, but computational efficiency decreases
Solution Approach 1:
The patent extracts and removes deep neural networks from the HDR imaging pipeline. Instead of using complex CNN-based approaches, the invention employs traditional signal processing techniques (wavelet transform) combined with simple frequency-domain operations, dramatically reducing computational requirements while maintaining effective HDR image generation.
Solution Approach 2:
The patent substitutes complex deep learning mechanisms with efficient mathematical transformations (wavelet transform and inverse wavelet transform). This replacement maintains the core functionality of processing multiple LDR images to generate HDR output while achieving significantly better computational efficiency and hardware compatibility.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Described herein is an image processing apparatus (701) comprising one or more processors (704) configured to: receive (601) a plurality of input images (301, 302, 303); for each input image, form (602) a set of decomposed data by decomposing the input image (301, 302, 303) or a filtered version thereof (307, 308, 309) into a plurality of frequency-specific components (313) each representing the occurrence of features of a respective frequency interval in the input image or the filtered version thereof; process (603) each set of decomposed data using one or more convolutional neural networks to form a combined image dataset (327); and subject (604) the combined image dataset (327) to a construction operation that is adapted for image construction from a plurality of frequency-specific components to thereby form an output image (333) representing a combination of the input images. The resulting HDR output image may have fewer artifacts and provide a better quality result. The apparatus is also computationally efficient, having a good balance between accuracy and efficiency.