Geophysical Texture Segmentation via Double-Windowed Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for texture segmentation in geophysical data, such as seismic data, often require training data, pre-defined similarity measures, or a well-known structure, which limits their effectiveness in complex datasets.
Innovation Solution
A computer-implemented method using double-windowed clustering analysis that extracts statistical distributions from user-defined windows, clustering them based on a standard similarity metric, without the need for pre-defined criteria, employing a new distance metric for probability distributions to identify similar textures.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If training data or pre-defined similarity measures are used for texture segmentation, then segmentation can be performed with known structures, but the method becomes ineffective for complex datasets without well-known structures
Solution Approach 1:
The method performs self-service by automatically learning texture characteristics directly from the input data without requiring external training data or pre-defined similarity measures. The algorithm extracts statistical features and computes similarity metrics autonomously, enabling it to adapt to any dataset structure independently.
Solution Approach 2:
The invention changes parameters by computing statistical distributions (mean, variance, skewness, kurtosis) of texture features dynamically from the data itself. These computed parameters serve as the basis for similarity measurement, allowing the method to adapt to different dataset structures by changing the statistical parameters rather than relying on fixed pre-defined measures.
2Reliability
If standard clustering methods are used without pre-defined criteria, then the method becomes more sensitive and robust, but requires new approaches for measuring similarity between probability distributions
Solution Approach 1:
The invention introduces an intermediary approach by using statistical distributions as mediators between raw texture data and clustering algorithms. Instead of directly comparing raw pixel values or using complex pre-defined similarity measures, the method computes statistical distributions (mean, variance, skewness, kurtosis) that serve as intermediate representations, simplifying the clustering process while improving robustness.
Solution Approach 2:
The method substitutes mechanical or pre-programmed similarity measures with a statistical-based system. Instead of using fixed thresholds or hand-crafted features, the invention replaces these mechanical approaches with dynamic statistical computations that adapt to the data distribution, thereby improving sensitivity and robustness.
3Productivity
If pre-defined texture attributes are designed to highlight different textures, then segmentation can be performed efficiently, but the method requires a priori knowledge of texture characteristics
Solution Approach 1:
The method performs preliminary action by automatically computing statistical distributions of texture features from the raw data before clustering. This preliminary statistical analysis replaces the need for a priori knowledge of texture characteristics, as the algorithm prepares the data by extracting mean, variance, skewness, and kurtosis values that capture essential texture properties without requiring pre-defined attributes.
Solution Approach 2:
The system performs self-service by automatically learning and extracting texture characteristics directly from the input data without requiring external training or pre-defined attributes. The algorithm independently computes statistical features and uses them for similarity measurement, making the segmentation process self-sufficient and adaptable to any dataset.
Data Source
AI summary
An automated method for texture segmentation (11) of geophysical data volumes, where texture is defined by double-window statistics of data values, the statistics being generated by a smaller pattern window moving around within a larger sampling window (12). A measure of “distance” between two locations is selected based on similarity between the double-window statistics from sampling windows centered at the two locations (13). Clustering of locations is then based on distance proximity (14).