Blind Deblurring via Internal Patch Recurrence
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing super-resolution and deblurring algorithms rely on assumptions about the blur kernel, which can lead to inadequate results when the actual kernel deviates from the assumed values, particularly in cases of camera shake, defocus, or low-grade optics.
Innovation Solution
The method utilizes the recurrence of small image patches across scales to estimate the optimal blur kernel, maximizing patch similarity and recovering the underlying unknown blur kernel without prior knowledge, enabling blind super-resolution and deblurring.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional super-resolution and deblurring algorithms use assumed blur kernels (such as PSF or standard LPF), then the algorithms can operate with simple models and assumptions, but the results become inadequate when the actual blur kernel deviates from the assumed values
Solution Approach 1:
The algorithm recovers the blur kernel directly from the input blurry image itself, without requiring external reference images or manual intervention. The patch recurrence property within the image provides the necessary information to self-determine the correct blur kernel, making the system autonomous and eliminating the need for external calibration data
Solution Approach 2:
The patent introduces an intermediary optimization process that searches for the blur kernel which maximizes patch recurrence across scales. This intermediary objective function serves as a bridge between the blurry image and the unknown sharp image, allowing the algorithm to indirectly recover the blur kernel by finding the transformation that best preserves self-similarity patterns
2Adaptability or versatility
If deblurring methods rely on prior knowledge assumptions about image structure (such as sparsity of gradients or polynomial spectrum decay), then the algorithms can be formulated with simple constraints, but the results fail when these assumptions do not hold for the actual image
Solution Approach 1:
The algorithm exploits the natural self-similarity property where small image patches recur across different scales in natural images. By copying and comparing patches at multiple scales, the method captures essential image structure without requiring explicit parametric assumptions about gradients or spectra, thereby preserving more authentic image details
Solution Approach 2:
The patent transforms the deblurring problem from directly optimizing image pixels under parametric assumptions to optimizing the blur kernel parameters that maximize patch recurrence. This parameter transformation allows the algorithm to adapt to different image contents by adjusting the blur kernel rather than forcing the image to conform to fixed parametric models
3Reliability
If the blur kernel is assumed to be a standard Low-Pass Filter such as Gaussian or bicubic kernel, then the algorithm implementation becomes straightforward, but the results are inadequate for real-world scenarios with camera shake, defocus, or low-grade optics
Solution Approach 1:
The algorithm transitions from using fixed, static blur kernel models to dynamically recovering the blur kernel specific to each input image. The blur kernel is not predetermined but adaptively determined based on the actual degradation present in the image, allowing the system to handle diverse real-world conditions including camera shake, defocus, and low-grade optics
Solution Approach 2:
The patent performs preliminary blur kernel recovery from the blurry image itself before executing the main deblurring operation. This preliminary action of kernel estimation from patch recurrence analysis prepares the correct degradation model in advance, ensuring that the subsequent deblurring process uses accurate kernel information rather than incorrect standard assumptions
Data Source
AI summary
Devices, systems, and methods of blind deblurring and blind super-resolution utilizing internal patch recurrence. Small signal patches tend to repeat “as is” across multiple scales of a natural signal. This fractal-like behavior is utilized for signal processing tasks, including “Blind Deblurring” or “Blind Super-Resolution”, namely, removing signal blur or increasing signal resolution without a-priori knowledge of the underlying blur kernel. While the cross-scale patch recurrence is strong in signals taken under ideal conditions, the cross-scale patch recurrence significantly diminishes when the acquisition blur deviates from an ideal blur. These deviations from ideal patch recurrences are used for recovering the underlying (unknown) blur kernel. The correct blur kernel is recovered by seeking the kernel which maximizes the patch similarity across scales of a related “reference” signal. For example, this reference signal may be the low-resolution input signal, the sharp deblurred-version of a blurry input signal, or the like. Quantitative and qualitative experiments indicate that this approach yields improved or superior results, in “Blind Deblurring” and in “Blind Super-Resolution”.