Multi-Frame Super-Resolution Network for Digital Zooming

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing digital zooming methods in image processing, particularly those using convolutional neural networks, often produce fuzzy outputs due to simulating image degradation and lack effective noise elimination, resulting in poor digital zooming and denoising capabilities.

Innovation Solution

A deep learning network training method that aligns and processes multiple frames of long-distance shot images with a short-distance standard image to enhance digital zooming and denoising, using a multi-frame super-resolution network model to improve image clarity and reduce noise.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Device complexity

If digital zooming is performed through simple upward sampling of clipping area, then device complexity is reduced, but image quality deteriorates producing fuzzy output

Engineering Contradiction:
Improvezooming process complexityVSAvoidimage quality
Core Design Contradiction:
Device complexityVSManufacturing precision

Solution Approach 1:

The patent applies preliminary action by capturing multiple frames of images at the same scene before digital zooming processing. These pre-captured frames are then aligned and fused to obtain a high-definition target image, thereby improving image quality without requiring complex optical zooming mechanisms.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent merges multiple low-resolution image frames through alignment and fusion processes to generate a single high-resolution image. This combining approach allows digital zooming to achieve optical-quality results without the complexity of physical zoom lenses.

Inventive Principle:
Principle #5Merging (Combining)

2Speed

If conventional digital zooming methods are used, then processing speed is maintained, but noise elimination capability is insufficient

Engineering Contradiction:
Improveprocessing speedVSAvoidnoise in image
Core Design Contradiction:
SpeedVSObject-generated harmful factors

Solution Approach 1:

The patent performs preliminary alignment and fusion of multiple frames before the final zooming operation. This preliminary processing removes noise and aligns image data, allowing subsequent processing to maintain high speed while producing clean, high-quality output images.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent uses multiple copies (frames) of the same scene captured at different times. By aligning and fusing these copies, the system eliminates noise through statistical averaging while preserving the underlying image structure, achieving both speed and quality.

Inventive Principle:
Principle #26Copying

3Manufacturing precision

If optical zoom lens is used, then image quality is maintained at high level, but device weight and cost increase

Engineering Contradiction:
Improveimage qualityVSAvoidzoom lens weight
Core Design Contradiction:
Manufacturing precisionVSWeight of moving object

Solution Approach 1:

The patent replaces the mechanical optical zooming system with a computational approach using multiple frame alignment and fusion algorithms. This substitution eliminates the need for heavy zoom lenses while achieving comparable or superior image quality through digital processing.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

Instead of using optical copies through lens systems, the patent uses digital copies (multiple captured frames) of the same scene. By processing these digital copies through alignment and fusion, the system achieves high-quality zooming without any physical optical components.

Inventive Principle:
Principle #26Copying

4Manufacturing precision

If multiple frames are aligned and fused, then image clarity is improved, but processing time increases

Engineering Contradiction:
Improveimage clarityVSAvoidprocessing time
Core Design Contradiction:
Manufacturing precisionVSLoss of time

Solution Approach 1:

The patent performs preliminary alignment and feature extraction from multiple frames before the final fusion step. This preliminary organization of data accelerates the subsequent fusion process, reducing overall processing time while maintaining high image clarity through thorough multi-frame analysis.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3816927B1Method and apparatus for training image processing models, and storage medium
Publication Date: 2024.04.24 BEIJING XIAOMI INTELLIGENT TECH CO LTD
  • EP3816927B1 patent drawingFigure 1
  • EP3816927B1 patent drawingFigure 2~4
  • EP3816927B1 patent drawingFigure 5~7

AI summary

Provided are a method and apparatus for training an image processing model, and a storage medium. The method includes: acquiring M frames of long-distance shot images of each of N training scenes respectively, both N and M being positive integers greater than 1; acquiring a short-distance standard image of each training scene; for each training scene, clipping the M frames of long-distance shot images to obtain a field of view the same as that of the short-distance standard image, and aligning the M frames of long-distance shot images after being clipped with the short-distance standard image; forming a training sample couple of each training scene by the M frames of long-distance shot images and the short-distance standard image after alignment, and forming a training sample set by N training sample couples corresponding to the N training scenes; and training an image processing model based on the training sample set.