Image Recognition Model Training via Localized Pixel Disturbance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Image recognition models trained with standard samples are prone to reduced accuracy when faced with increased image interference, compromising security in applications like face recognition payment authentication.
Innovation Solution
An enhanced training method is introduced, where disturbance values are added to image samples based on a predetermined distribution centered around a reference pixel, creating extended samples that are used to train the image recognition model, thereby improving its robustness against interference.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If an image recognition model is trained by using only a standard training sample, then the training process is simple and fast, but accuracy of an image recognition result is reduced when image interference increases
Solution Approach 1:
The patent applies preliminary action by pre-processing training samples to add disturbance values before training the model. The disturbance values are generated and integrated into the training data in advance, allowing the model to learn from perturbed samples without increasing the complexity of the training algorithm itself. This prepares the model beforehand to handle interference in real-world applications.
Solution Approach 2:
The patent changes parameters by introducing disturbance values that modify the pixel values in training images. These disturbance values are added to the image pixels according to specific distribution patterns, effectively transforming the training samples to include various levels of interference. This parameter modification allows the model to become robust against interference without complicating the training process.
2Reliability
If disturbance values are added to training samples to improve robustness, then reliability of the model improves, but the training process becomes more complex
Solution Approach 1:
The patent applies local quality by adding disturbance values with different characteristics to different regions of the training images. The disturbance values are generated based on local properties such as pixel intensity and spatial position, creating non-uniform perturbations that mimic real-world interference patterns. This localized approach improves robustness without requiring complex global transformations.
Solution Approach 2:
The patent uses copying by generating multiple versions of each training sample with different disturbance values applied. Instead of modifying the original training data structure, the system creates copies of the samples with varied perturbations, allowing the model to learn from multiple variations without increasing the fundamental training process complexity.
3Reliability
If extended samples with disturbance values are used for training, then the model maintains accurate recognition under interference, but more processing time is required during training
Solution Approach 1:
The patent applies partial action by adding disturbance values to only a portion of the training samples or to specific regions within samples, rather than uniformly processing every pixel of every image. This selective approach maintains recognition accuracy under interference while reducing the overall computational burden and training time compared to exhaustive processing.
Data Source
AI summary
Implementations of the present specification provide an enhanced training method for an image recognition model. A predetermined quantity or predetermined proportion of samples are randomly selected from a first sample set as a seed sample for extension to obtain several extended samples. The extended sample is obtained by adding disturbance to an original image without changing an annotation result. In a sample extension process, disturbance values are arranged towards neighborhood in predetermined distribution with a reference pixel as a reference, so that real disturbance can be well simulated. Because the annotation result of the extended sample remains unchanged after the disturbance is added, an image recognition model trained by using the extended sample can well recognize a target recognition result of an original image, thereby improving robustness of the image recognition model.