Signal Processing Using Dynamic Reference Pixel Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing pattern identification methods, such as LGBP, require multiple Gabor Wavelet filters for high accuracy, resulting in large data amounts and increased hardware costs, while methods like incremental encoding reduce data but compromise identification accuracy.
Innovation Solution
A signal processing method that applies spatial filtering using Gabor Wavelet filters with biased frequency responses and dynamically adjusts the relative position of reference pixels based on the frequency responses or machine learning to optimize feature extraction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple Gabor Wavelet filters are used for high identification accuracy, then pattern identification accuracy is improved, but data amount increases and hardware cost increases
Solution Approach 1:
The patent extracts only the most essential frequency components from the filtering result using a selectively determined reference pixel, rather than processing all frequency components equally. This extraction approach reduces data amount while preserving the most discriminative features for pattern identification.
Solution Approach 2:
The patent applies different processing strategies to different frequency components by dynamically selecting reference pixels based on frequency response characteristics. This local quality approach ensures that each frequency component is processed according to its specific properties, maintaining identification accuracy while reducing overall data volume.
2Measurement precision
If multiple Gabor Wavelet filters are used for high identification accuracy, then pattern identification accuracy is improved, but hardware cost increases
Solution Approach 1:
The patent extracts only the most essential frequency components from the filtering result using a selectively determined reference pixel, rather than processing all frequency components equally. This extraction approach reduces data amount while preserving the most discriminative features for pattern identification.
Solution Approach 2:
The patent applies different processing strategies to different frequency components by dynamically selecting reference pixels based on frequency response characteristics. This local quality approach ensures that each frequency component is processed according to its specific properties, maintaining identification accuracy while reducing overall data volume.
3Quantity of substance
If incremental encoding is used to reduce data amount, then data amount is reduced, but pattern identification accuracy deteriorates
Solution Approach 1:
The patent changes the encoding parameters by dynamically selecting reference pixels based on frequency response characteristics or machine learning results. This parameter change allows the encoding process to adapt to different frequency components, preserving identification accuracy while achieving data reduction.
Solution Approach 2:
The patent introduces dynamic reference pixel selection that adapts to the specific characteristics of each frequency component. This dynamics approach allows the system to optimize the encoding process for each component, maintaining accuracy while reducing data volume, unlike static incremental encoding.
Data Source
AI summary
There is provided with a signal processing method. A filtering result is generated by performing spatial filtering on multi-dimensional data. Encoding result data is output by encoding the filtering result using a value at a pixel of interest of the filtering result and a value at a reference pixel located at a relative position with respect to the pixel of interest. The relative position of the reference pixel is decided in advance according to a characteristic of a spatial filter used in the spatial filtering step.