Image Upscaling via Segmented Machine Learning Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for generating upscaled images using deep learning models require a large computational load due to the need for numerous filter convolutions, especially when generating images with a large number of channels.
Innovation Solution
The proposed method involves using two distinct machine learning models to generate upscaled images with different effects. The first model generates a high-resolution image with many high-frequency components, and the second model generates a lower-resolution image with fewer high-frequency components. These images are then combined using a weighted average to produce a final upscaled image, thereby reducing the computational load.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If images are generated from a feature map with a large number of channels using a deep learning model, then two upscaled images with different effects can be generated, but the computational load increases significantly
Solution Approach 1:
The patent segments the deep learning model into multiple independent machine learning models, each responsible for generating one upscaled image with a specific effect. This segmentation allows the system to process images with fewer channels at a time, reducing the computational load while maintaining the ability to generate multiple images with different effects
Solution Approach 2:
The patent changes the processing approach by working with images having fewer channels instead of processing the full high-channel feature map directly. This dimensional reduction in the input space allows for more efficient computation while still achieving the desired multiple output effects through separate model processing
2Manufacturing precision
If a deep learning model processes a feature map with many channels to generate upscaled images, then image quality can be maintained, but the number of filter convolutions and computational requirements increase
Solution Approach 1:
The patent divides the complex deep learning model into multiple simpler machine learning models that process images with fewer channels. Each model handles a specific aspect of image upscaling with reduced filter convolution requirements, thereby maintaining image quality while reducing overall computational complexity
Solution Approach 2:
The patent introduces intermediate processing steps where images with fewer channels are generated first, then processed by separate machine learning models. This intermediary approach reduces the direct computational burden on any single model while preserving the ability to produce high-quality output images
Data Source
AI summary
An image processing method includes generating a first image by inputting an input image or an image based on the input image into a first machine learning model, generating a second image by inputting the first image into a second machine learning model different from the first machine learning model, and generating a third image using the first image and the second image. Each of the first image, the second image, and the third image has a larger number of pixels than those of the input image. The second image has fewer high-frequency components than those of the first image.