Diffusion Image Restoration Using Shared Initial Denoising Steps

Solution Overview

Problem

Deep learning-based image restoration models require a large number of processing steps for each image in a sequence, leading to increased processing time and computational demands, particularly in multiphasic imaging where large volumes of images are acquired over time or space.

Innovation Solution

A method utilizing a shared phase representative image to reduce the number of diffusion steps by first denoising a series of images using this representative image, followed by additional denoising steps specific to each image, thereby reducing the total number of steps required for image restoration.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Manufacturing precision

If a deep learning model processes each image in a sequence by repeating the full series of denoising steps, then each image can be restored with high quality, but the processing time and computational capacity required increase significantly

Engineering Contradiction:

Improveimage restoration qualityVSAvoidprocessing time

Core Design Contradiction:

Manufacturing precisionVSLoss of time

Solution Approach 1:

The patent performs preliminary denoising steps using a shared representative image that captures common features across multiple images. By pre-processing the shared content in the first sequence of T1 steps, the method avoids repeating these steps for each individual image, thereby reducing total processing time while maintaining restoration quality.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent merges multiple images into a single shared representative image that contains common structural and semantic information. This combined representation is then processed once through the diffusion model, and the results are reused across all source images, eliminating redundant computations and reducing overall processing time.

Inventive Principle:

Principle #5Merging (Combining)

2Manufacturing precision

If a deep learning model processes each image in a sequence by repeating the full series of denoising steps, then each image can be restored with high quality, but the computational capacity required increases significantly

Engineering Contradiction:

Improveimage restoration qualityVSAvoidcomputational capacity

Core Design Contradiction:

Manufacturing precisionVSPower

Solution Approach 1:

The patent performs preliminary denoising steps using a shared representative image that captures common features across multiple images. By pre-processing the shared content in the first sequence of T1 steps, the method avoids repeating these steps for each individual image, thereby reducing total processing time while maintaining restoration quality.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent merges multiple images into a single shared representative image that contains common structural and semantic information. This combined representation is then processed once through the diffusion model, and the results are reused across all source images, eliminating redundant computations and reducing overall processing time.

Inventive Principle:

Principle #5Merging (Combining)

3Productivity

If the number of denoising steps is reduced to decrease processing time, then processing efficiency improves, but image restoration quality may deteriorate

Engineering Contradiction:

Improveprocessing efficiencyVSAvoidimage restoration quality

Core Design Contradiction:

ProductivityVSManufacturing precision

Solution Approach 1:

The patent applies different numbers of denoising steps to different parts of the processing workflow. The first sequence of T1 steps processes the shared representative image to capture global structures, while the second sequence of T2 steps processes each individual image to refine local details. This differentiated approach ensures high restoration quality while improving overall efficiency.

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

The patent changes the parameter of denoising step count based on the processing stage and image type. Using fewer steps (T1) for the shared representative image and more steps (T2) for individual images allows the system to optimize the balance between processing efficiency and restoration quality dynamically.

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS20260044938A1Fast diffusion-based image restoration workflow via sharing of initial diffusion steps

Publication Date: 2026.02.12 CANON MEDICAL SYST CORP

AI summary

A method includes obtaining a diffusion-based probabilistic model to perform denoising over T steps; determining a shared phase representative image based on a plurality of phase images; generating a sequence of representative images by performing a first sequence of T1 denoising sampling steps using the obtained model starting with a start image and the shared phase representative image as initial first sequence inputs; determining, from the generated sequence of representative images, an intermediate image; and for each phase image in the plurality of phase images: generating a corresponding sequence of restored images by performing a second sequence of T2 denoising sampling steps using the obtained model with the intermediate image and the input image as initial second sequence inputs; and determining a corresponding final restored image for each phase image based on the corresponding sequence of restored images, wherein T1 and T2 are integers greater than or equal to 1.