Deep Neural Network Image Super-Resolution via Nonlinear Convolution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing methods for generating high-resolution pictures from low-resolution images are inefficient, leading to high interaction costs and low interaction efficiency due to the need for large storage media and high-speed broadband for multimedia information exchange.
Innovation Solution
A method using a deep neural network model with non-linear conversion convolution layers to convert low-resolution pictures into high-resolution pictures, reducing storage and bandwidth requirements by acquiring and determining the appropriate deep neural network model for image enhancement.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high-resolution multimedia information is transmitted and stored using traditional methods, then picture quality is maintained, but storage media requirements increase and bandwidth consumption increases
Solution Approach 1:
The patent uses deep learning models to create a computational copy of the high-resolution image reconstruction process. Instead of physically storing and transmitting large amounts of high-resolution data, the system trains a neural network model on high-resolution images and then uses this model to generate high-resolution outputs from low-resolution inputs, effectively copying the essential visual information in a compressed form.
Solution Approach 2:
The patent transforms the problem from physical parameter space (image resolution, storage capacity, bandwidth) to computational parameter space (neural network architecture, learning rates, loss functions). By changing the parameters of the deep learning model during training and inference, the system can adapt to different resolution requirements without physically altering storage or transmission infrastructure.
2Measurement precision
If high-resolution multimedia information is transmitted and stored using traditional methods, then picture quality is maintained, but interaction efficiency decreases
Solution Approach 1:
The patent performs preliminary action by pre-training the deep learning model on large datasets of high-resolution images before actual use. This pre-processing step captures the essential patterns and features of high-resolution imagery, enabling the model to rapidly reconstruct high-quality images from low-resolution inputs during interaction, thereby improving interaction efficiency.
3Measurement precision
If deep neural network models are used to convert low-resolution pictures to high-resolution pictures, then conversion accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent segments the deep neural network into multiple specialized components: convolutional layers for feature extraction, activation layers for non-linear transformation, and pooling layers for dimensionality reduction. This segmentation allows each component to focus on specific aspects of image reconstruction, improving overall accuracy while enabling optimized computation for each segment.
Solution Approach 2:
The patent introduces multiple dimensions to the computational process by adding depth (multiple layers), width (multiple filters/channels), and temporal dimensions (sequential processing steps). This multi-dimensional approach allows the model to capture complex image features and relationships that would be impossible with simple two-dimensional convolution, thereby improving conversion accuracy while distributing computational load across different dimensions.
Data Source
AI summary
This application provides a method for generating a high-resolution picture performed by a computer device. The method includes: acquiring at least one deep neural network model; acquiring a low-resolution picture; determining a corresponding deep neural network model according to the low-resolution picture; and converting the low-resolution pictures into a high-resolution picture through the deep neural network model, the deep neural network model including a plurality of non-linear conversion convolution layers that alternately use different parameter matrices as convolution template parameters.