Object Classification Using Region and Partial-Region Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image classification methods struggle to accurately classify objects of varying sizes and shapes due to the limitations of using fixed-sized low-magnification images, which can either truncate large objects or include irrelevant background information, leading to inaccurate region information extraction.
Innovation Solution
An object classification device that utilizes an arithmetic operation device to determine object regions, generate unit feature values, and classify objects based on region and partial region feature values, employing methods like U-Net and Convolutional Neural Networks to handle objects of different sizes and shapes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the size of the low-magnification image is determined according to the small object, then the small object can be fully captured, but the large object cannot be completely fit in the patch
Solution Approach 1:
The patent segments the image processing task into two stages: first performing broad classification using a fixed-sized low-magnification image to identify object locations, then performing local classification using high-magnification images of specific regions to accurately extract features of individual objects regardless of their size. This segmentation allows both small and large objects to be properly processed in their respective optimal viewing conditions.
2Measurement precision
If the size of the low-magnification image is determined according to the large object, then the large object can be fully captured, but many other objects or dust may be contained around the small object
Solution Approach 1:
The patent extracts only the necessary region information for classification by using the fixed-sized low-magnification image to identify object locations, then extracting only the relevant high-magnification regions for each detected object. This extraction approach obtains sufficient information for classification while excluding irrelevant surrounding objects and dust that would interfere with the classification process.
3Productivity
If fixed-sized low-magnification images are used for classification, then processing efficiency is improved, but accurate extraction of region information for objects of varying sizes becomes impossible
Solution Approach 1:
The patent divides the classification process into two efficient stages: first using fixed-sized low-magnification images for rapid broad classification to identify object locations, then using targeted high-magnification images only for the detected objects to extract precise region information. This segmented approach maintains processing efficiency while achieving accurate extraction for objects of varying sizes.
Data Source
AI summary
An object classification device generates region unit information indicating the region information of the object, classifies the object by determination of the object type or state based on the region unit feature value extracted from the region unit information, and the partial region unit feature value extracted from the partial region of the input image, and displays the object classification result for the user. This makes it possible to classify the object type or state accurately even in spite of co-existence of the objects different in size, length, shape, and the like in the image.