Image Recognition Database Enhancement via Incomplete Feature Blocks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Image recognition technologies, such as those using convolution neural networks, face challenges in accuracy when a shielding object covers the target, leading to reduced recognition rates.
Innovation Solution
An image recognition method and device that trains a learning database using feature blocks of a target image, randomly selects incomplete feature blocks, and adds them to form an enhancement learning database to improve recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a shielding object covers the target in the image, then the recognition system may still attempt to recognize the target, but the recognition accuracy deteriorates
Solution Approach 1:
The patent segments the image into multiple feature blocks and processes them independently through the neural network. By dividing the image into smaller regions, the system can identify recognizable features in unshielded areas while ignoring or weighting shielded areas differently, thereby maintaining recognition accuracy even when parts of the target are covered
Solution Approach 2:
The patent dynamically adjusts the weighting parameters assigned to different feature blocks based on their recognition confidence levels. Feature blocks with higher confidence (likely unshielded or less affected areas) receive higher weights, while low-confidence blocks (likely shielded areas) receive lower weights. This parameter adjustment allows the system to adapt to shielding conditions and maintain accurate recognition
2Reliability
If the learning database is trained only with complete feature blocks, then the training process is simpler, but the recognition performance deteriorates when shielding objects are present
Solution Approach 1:
The patent performs preliminary action by pre-processing images to generate multiple feature blocks with different weighting schemes before actual recognition. During training, the system pre-establishes the framework for generating confidence levels and weight assignments for each feature block. This preliminary preparation enables the system to handle shielding scenarios effectively without requiring complex real-time adjustments during recognition
Solution Approach 2:
The patent introduces dynamics into the learning database by implementing adaptive weighting mechanisms where feature block weights are not fixed but dynamically adjusted based on recognition confidence. The system can adaptively reweight feature blocks during both training and recognition phases, allowing the learning database to flexibly respond to varying shielding conditions rather than relying on static complete-feature-block training
Data Source
AI summary
An image recognition method adapted to recognize a target image is provided. The method includes: providing the target image; training a learning database based on a plurality of feature blocks of the target image; randomly obtaining a plurality of incomplete feature blocks of the target image; adding the plurality of incomplete feature blocks into the learning database to form an enhancement learning database; and recognizing the target image based on the enhancement learning database. In addition, an image recognition device is also provided.