Image Analysis Device for Generalization Object Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image recognition systems using convolutional neural networks (CNNs) face difficulties in recognizing unknown objects that are not registered in the learning dataset, limiting their ability to generalize beyond known objects.
Innovation Solution
An image analysis device and method that calculates feature amount information from input images, recognizes known objects, and identifies generalization objects by combining features of known objects, allowing for the recognition of unknown objects without requiring changes to the learning data or network size, using a configuration with a receiver, calculator, known-object recognizer, generalization-object recognizer, and output controller.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If conventional CNN-based image recognition is used, then known objects registered in learning dataset can be recognized, but unknown objects cannot be recognized
Solution Approach 1:
The patent segments object recognition into two independent processes: known object recognition using CNN and unknown object recognition through shape analysis. This segmentation allows each process to specialize - CNN handles familiar objects while shape analysis handles novel objects, resolving the contradiction between adaptability and reliability
Solution Approach 2:
The patent introduces shape information as an intermediary element that bridges known and unknown objects. By extracting and comparing shape features independently of object category, the system can recognize unknown objects through shape similarity to known objects, enabling generalization without compromising recognition reliability
2Adaptability or versatility
If learning dataset is expanded to include more object categories, then more objects can be recognized, but system complexity and data requirements increase
Solution Approach 1:
The patent extracts shape information as a separate, category-independent feature from the complex object recognition process. By taking out shape analysis from the CNN-based category-specific recognition, the system achieves universal object recognition without expanding learning datasets or increasing model complexity
Solution Approach 2:
The patent creates a universal shape analysis component that serves multiple functions: it complements CNN for known object recognition and enables independent unknown object recognition. This multi-functional approach expands object category coverage without requiring separate systems for each function, avoiding complexity multiplication
Data Source
AI summary
According to one embodiment, an image analysis device includes one or more processors configured to receive input of an image; calculate feature amount information indicating a feature of a region of the image; recognize a known object from the image on the basis of the feature amount information, the known object being registered in learning data of image recognition; recognize a generalization object from the image on the basis of the feature amount information, the generalization object being generalizable from the known object; and output output information on an object identified from the image as the known object or the generalization object.