Super-Resolution CNN Digital Zoom Image Definition

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional digital zoom methods in mobile terminals result in image quality deterioration, characterized by decreased definition and noise amplification, especially when using low-definition training data.

Innovation Solution

An image processing method utilizing a target super-resolution convolutional neural network model, trained with high-definition and low-definition training images, along with a mask image, to improve image quality by adjusting the model based on loss cost results and applying it for digital zoom, thereby enhancing the definition of zoomed images.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Manufacturing precision

If conventional digital zoom uses low-definition training data for neural network training, then the processing speed and resource consumption are reduced, but the image definition decreases and noise is amplified

Engineering Contradiction:

Improveimage definitionVSAvoidnoise amplification

Core Design Contradiction:

Manufacturing precisionVSObject-generated harmful factors

Solution Approach 1:

The patent segments the training image into multiple patches, where each patch is processed independently by the neural network. This segmentation allows the model to focus on local features and textures, improving image definition while reducing noise amplification in each localized region.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent applies different processing strategies to different regions of the image. By training the neural network on patches with various characteristics (edges, textures, smooth regions), the model learns to preserve local quality and reduce noise amplification in specific areas while maintaining overall image definition.

Inventive Principle:

Principle #3Local quality

2Adaptability or versatility

If conventional digital zoom is applied to mobile terminals, then the zoom function is enabled, but image quality deteriorates with decreased definition and noise amplification

Engineering Contradiction:

Improvezoom functionVSAvoidimage quality

Core Design Contradiction:

Adaptability or versatilityVSManufacturing precision

Solution Approach 1:

The patent performs preliminary training of the neural network model using high-definition training images before deployment. This preliminary action prepares the model to handle various image scenarios, enabling it to maintain image quality when zoom functions are applied to mobile terminals.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The neural network model serves as an intermediary between the low-definition captured image and the final zoomed output. It processes the input image through multiple convolutional layers, acting as a mediator that enhances definition and reduces noise before producing the final zoomed image.

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS11704771B2Training super-resolution convolutional neural network model using a high-definition training image, a low-definition training image, and a mask image

Publication Date: 2023.07.18 HUAWEI TECH CO LTD

AI summary

An image processing method and a device, where the image processing method is performed by a terminal having a digital zoom function, and the method includes determining a target zoom magnification based on a selection input of a user, collecting a to-be-processed image, and processing the to-be-processed image using a target super-resolution convolutional neural network model to obtain a processed image corresponding to the target zoom magnification, where the target super-resolution convolutional neural network model is obtained by training a super-resolution convolutional neural network model using a high-definition training image, a low-definition training image, and a mask image.