Probabilistic Image Feature Extraction for Noise Robustness
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image processing technologies face challenges in effectively extracting and recognizing image features, particularly in varying conditions, as they rely on fixed methods like SIFT, HOG, LBP, and MCT, which may not adapt well to noise and changes in pixel values.
Innovation Solution
An image feature extraction method using a probabilistic model based on pixel values within a kernel, where the model determines whether to change image feature information of a pixel by comparing a probability value to a random value, allowing for adaptive feature extraction and recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If fixed image feature extraction methods (SIFT, HOG, LBP, MCT) are used, then the extraction process is simple and fast, but the recognition rate decreases and robustness to noise deteriorates under varying image conditions
Solution Approach 1:
The patent applies dynamics by transitioning from fixed, deterministic image feature extraction methods to a dynamic probabilistic approach. The system uses probability distribution functions to model pixel values and dynamically adjusts feature extraction based on learned probability parameters from training data, enabling adaptation to varying image conditions while maintaining robustness against noise and transformations.
Solution Approach 2:
The patent implements parameter changes by learning probability parameters (mean and standard deviation) from training images and using these parameters to adjust the feature extraction process. The probabilistic model changes parameters based on the input image characteristics, allowing the system to adapt extraction behavior to different lighting, scale, and rotation conditions, thereby improving recognition rate without excessive complexity.
2Adaptability or versatility
If probabilistic models are used to adapt to varying conditions, then robustness to noise and changes improves, but the computational complexity and processing time increase
Solution Approach 1:
The patent applies preliminary action by pre-training the probabilistic model on a large dataset of images with varying conditions before actual feature extraction. The probability distribution parameters are learned in advance during training, so that during runtime, the system only needs to apply the pre-computed probabilistic model to new images, significantly reducing processing time while maintaining high adaptability to varying conditions.
3Productivity
If traditional fixed methods are used, then processing speed is fast, but the ability to handle noise and varying pixel values deteriorates
Solution Approach 1:
The patent replaces the mechanical, deterministic approach of traditional image feature extraction with a probabilistic model-based system. Instead of using fixed algorithms that mechanically compute features, the system uses probability distribution functions to model pixel values and their variations, substituting rigid mechanical processing with statistical reasoning that naturally handles noise and varying conditions while maintaining efficient processing speed.
Data Source
AI summary
At least one example embodiment discloses an image feature extracting method. The method includes determining a probabilistic model based on pixel values of pixels in a kernel, determining image feature information of a current pixel of the pixels in the kernel and determining whether to change the image feature information of the current pixel based on a random value and a probability value of the current pixel, the probability value being based on the probabilistic model.