Block-Selective Image Enhancement for Lower ML Memory Load
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing methods using machine learning models for image estimation are inefficient in terms of memory usage and processing load, particularly when only a specific region of an input image needs to be processed.
Innovation Solution
The method involves dividing the input image into blocks, determining a specific region of interest based on optical characteristics, and applying sharpening or resolution enhancement processing only to the relevant areas using a trained machine learning model, thereby reducing unnecessary processing by excluding areas outside the specific region.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If image estimation processing is performed on the entire input image using a machine learning model, then image quality improvement is achieved, but memory usage and processing load increase significantly
Solution Approach 1:
The input image is divided into multiple divided images (e.g., 4 quadrants) before processing. The machine learning model processes only the divided images that contain the specific region of interest, rather than the entire image. This segmentation reduces the amount of data processed and memory required while maintaining image quality in the target area.
Solution Approach 2:
The patent extracts only the necessary portion of the image (the specific region of interest) for processing by determining which divided images contain this region. Areas outside the specific region are excluded from processing, reducing memory usage and computational load while preserving quality where needed.
2Manufacturing precision
If image estimation processing is performed on the entire input image, then comprehensive image enhancement is achieved, but processing time and computational load increase
Solution Approach 1:
The image is segmented into multiple divided images, and processing is performed only on those containing the specific region of interest. This reduces the total number of processing operations required, thereby decreasing processing time and improving productivity while maintaining enhancement quality in the target area.
Solution Approach 2:
Instead of processing the entire image (excessive action), the patent applies processing only to the necessary partial areas containing the specific region of interest. This partial action reduces computational load and processing time while achieving the required image enhancement where needed.
3Stability of the object's composition
If processing is applied to the entire input image, then uniform image quality is maintained, but unnecessary processing increases memory consumption
Solution Approach 1:
The patent applies different processing strategies to different regions: the specific region of interest receives full machine learning-based estimation processing, while other regions are handled differently or not processed at all. This local quality approach maintains image quality consistency where needed while reducing overall memory consumption by excluding unnecessary processing areas.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An image processing method includes determining a specific region in a first image (S104), acquiring at least one second image corresponding to some of a plurality of partial images of the first image (S105), generating a third image corresponding to the second image by inputting the second image into a machine learning model (S106), and generating a fourth image based on the first image and the third image (S110), and the second image includes at least a part of the specific region.