Image Analysis Method Using Global and Local Feature Vector Merging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Convolutional neural networks face limitations when processing large medical images due to hardware constraints, as they exceed conventional network specifications.
Innovation Solution
An image analyzing method that extracts a first feature vector from global image information, divides the image into non-overlapping regions, inputs each region into a convolutional neural network to obtain second feature vectors, and merges these with the global feature vector to perform an image analysis process, using a global filter to transform feature maps into feature values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the entire large medical image is input into a convolutional neural network for analysis, then comprehensive feature extraction can be achieved, but hardware specifications are exceeded and the system cannot process the image
Solution Approach 1:
The large medical image is divided into multiple smaller regions or patches that can be individually processed by the convolutional neural network. Each region is analyzed separately and then the results are aggregated to form the final analysis, allowing the system to handle large images that exceed hardware constraints while maintaining comprehensive feature extraction capability
2Quantity of substance
If the image is divided into regions for processing, then hardware limitations are overcome, but global context information may be lost
Solution Approach 1:
After each image region is processed independently by the convolutional neural network to extract local features, the feature vectors from all regions are merged or aggregated together. This combining step reconstructs the global context information by integrating the local features, ensuring that both regional details and overall image context are preserved in the final analysis
Data Source
AI summary
An image analyzing method is provided and includes: extracting a first feature vector according to global information of a digital image; dividing the digital image into multiple regions, and inputting each region into a convolutional neural network to obtain a second feature vector; merging the first feature vector with the second feature vectors to obtain a third feature vector; and performing an image analyzing process according to the third feature vector.