Image Categorization Device Using Adaptive Model Creation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image indexing technologies are limited in categorizing objects specific to user data, as they rely on general object models, failing to detect undefined or user-specific objects effectively.
Innovation Solution
A data processing device that uses feature amounts to categorize objects by storing model data pieces with detection counts, judging non-categorizable data pieces, specifying common feature amounts, and creating new models based on these features to improve categorization accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If general object models are used for image categorization, then categorization speed is improved, but the ability to detect user-specific objects deteriorates
Solution Approach 1:
The patent segments the categorization process into two distinct phases: (1) rapid initial categorization using pre-built general object models, and (2) adaptive model creation for user-specific objects based on frequently detected features. This segmentation allows the system to maintain high speed for common objects while developing specialized detection capabilities for user-specific objects over time.
Solution Approach 2:
The patent performs preliminary actions by pre-building general object models before actual categorization tasks. These pre-built models enable fast initial categorization. Additionally, the system performs preliminary analysis of uncategorized images to identify frequently detected features, which are then used to create adaptive models for user-specific objects.
2Loss of time
If conventional image indexing techniques are used, then processing time is reduced, but categorization accuracy for undefined objects deteriorates
Solution Approach 1:
The patent implements a self-service mechanism where the system automatically analyzes uncategorized images, identifies frequently detected features, and creates its own adaptive models for user-specific objects without requiring manual intervention. This self-service approach maintains low processing time while progressively improving categorization accuracy for undefined objects.
Solution Approach 2:
The patent incorporates feedback loops where categorization results are continuously analyzed. When objects are repeatedly detected as uncategorized, the system feeds this information back into the model creation process, using the accumulated feature data to generate new adaptive models that improve future categorization accuracy.
3Device complexity
If general object models are used, then device complexity is reduced, but the ability to handle diverse objects deteriorates
Solution Approach 1:
The patent creates a dynamic model system where the collection of object models is not fixed but evolves over time. General object models provide a simple baseline, while adaptive models are dynamically created based on detected user-specific objects. The system automatically manages this dynamic model collection, adding new models only when needed, thus maintaining low complexity while improving versatility.
Solution Approach 2:
The patent creates a universal categorization framework that can handle both general objects (using pre-built models) and user-specific objects (using adaptive models). The same system infrastructure supports multiple model types, making the device universally applicable to diverse objects without requiring separate specialized systems for each object type.
Data Source
AI summary
A data processing device provides a result of categorization that is satisfactory to a user. The data processing device: stores model data pieces indicating detection counts of feature amounts; judges, for each target data piece, whether the target data piece is a non-categorization data piece including an uncategorizable object, using the model data pieces and the detection count of each of at least two feature amounts detected in the target data piece; when two or more of the target data pieces are judged to be non-categorization data pieces, specifies at least two feature amounts that are included and detected the same number of times, in a predetermined number or more of the non-categorization data pieces; and newly creates a model data piece based on the at least two specified feature amounts, using a class creation method, and stores the model data piece into the storage unit.