Dictionary Learning for Image Detection Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In image processing, computers face challenges in detecting specific targets from captured images due to low detection rates and positional inaccuracies when relying on existing reference data dictionaries.
Innovation Solution
A dictionary learning device and method that extracts and learns from multiple image areas, including reference, enlarged, and reduced image areas, to improve detection accuracy and positional certainty by generating more comprehensive teacher data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a computer uses reference data from a dictionary to detect targets in captured images, then the detection processing can be performed, but the detection rate is low and positional accuracy is poor
Solution Approach 1:
The patent divides the image area into multiple types (reference image area, enlarged image area, reduced image area) and extracts features from each segment separately. This segmentation allows the system to learn diverse characteristics of the detection target under different spatial contexts, improving both detection rate and positional accuracy by capturing local and global image properties.
Solution Approach 2:
The patent extends the feature extraction from a single reference image area to multiple dimensional variations including enlarged and reduced image areas. This multi-dimensional approach enriches the feature space by incorporating spatial scaling information, enabling the dictionary learning to capture target characteristics across different scales and improve detection reliability and precision.
2Device complexity
If the dictionary is learned using only reference image area, then the learning process is simple, but the detection reliability is insufficient
Solution Approach 1:
The patent performs preliminary extraction of multiple image areas (reference, enlarged, reduced) before the dictionary learning process. By pre-processing and organizing diverse image regions in advance, the system prepares comprehensive training data that enhances detection reliability without significantly increasing the complexity of the core learning algorithm.
Solution Approach 2:
The patent creates a multi-functional feature extraction system that processes different types of image areas (reference, enlarged, reduced) through a unified dictionary learning framework. This universal approach allows the same learning mechanism to handle diverse spatial information, improving detection reliability while maintaining learning process manageability.
Data Source
AI summary
A reference data extraction unit extracts, from a photographic image from an imaging device that captures an image of an object to be recognized, an image of a reference image region serving as a reference and containing a detection subject in the object. A expanded data extraction unit extracts from the photographic image an image of an expanded-image region, which is an image region that includes the reference image region and is larger than the reference image region. A reduced data extraction unit extracts from the photographic image an image of a reduced-image region, which is an image region that includes the detection subject and is smaller than the reference image region, with the result that a portion of the object is outside of the region. A learning unit uses the extracted images of the image region to learn a dictionary.