Parallel Hilbert-Huang Transform for Real-Time 2D Image Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The Hilbert Huang Transform (HHT) faces high computational complexity and boundary effect issues when processing large 2D images with non-linear and non-stationary data, limiting its practical application due to inefficient envelope construction and repeated stoppage criteria tests.
Innovation Solution
Implementing a multi-thread processing unit to simultaneously construct local lower and upper envelopes and determine the number of Empirical Mode Decomposition (EMD) interactions, reducing computational complexity and avoiding boundary effects by mapping each data point to a thread level and using GPUs for parallel processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional sequential processing is used to construct envelopes and perform EMD iterations, then computational accuracy is maintained, but processing time becomes excessively long for large 2D images
Solution Approach 1:
The patent divides the 2D image into multiple independent data points that can be processed simultaneously by different threads. Each data point is assigned to a separate computational thread, allowing parallel construction of upper and lower envelopes across the entire image, thereby reducing processing time while maintaining accuracy through independent computation of each segment
Solution Approach 2:
The patent transitions from sequential one-dimensional processing to parallel multi-dimensional processing by utilizing the thread-level parallelism capability of GPUs. Multiple threads operate simultaneously on different data points across the 2D image space, effectively adding a parallelism dimension to the computation and dramatically reducing processing time
2Measurement precision
If multiple EMD interactions are performed to ensure accurate decomposition, then decomposition quality improves, but computational complexity increases significantly
Solution Approach 1:
The patent performs preliminary actions by simultaneously constructing upper and lower envelopes for all data points in parallel before initiating EMD iterations. This pre-computation of envelope information for all data points upfront reduces the computational complexity of subsequent iterations while ensuring accurate decomposition through complete envelope availability
Solution Approach 2:
Each computational thread independently determines its own stoppage criteria based on local envelope construction results without requiring global coordination. This self-service approach allows each thread to autonomously decide when to stop EMD iterations for its assigned data points, reducing overall computational complexity while maintaining decomposition quality through localized optimization
3Ease of operation
If conventional CPU processing is used, then program control is simple, but processing speed is insufficient for real-time application of large 2D images
Solution Approach 1:
The patent replaces the traditional CPU-based sequential processing mechanism with GPU-based parallel processing. This substitution leverages the massive parallel computation capability of GPUs to achieve real-time processing speeds for large 2D images while maintaining relatively simple program control through standard CUDA or OpenCL programming models
Data Source
AI summary
An apparatus, computer program product and method of analyzing two-dimensional data input. The system is known as Syneren Signal and Image Enhancement Technology (SIETECH). SIETECH can be implemented in software or in Field Programmable Gate Array (FPGA) hardware. Some embodiments of the present invention pertain to apparatuses, method, and a computer program that is configured to cause the central processor to pass the input data to the multi-thread processors, wherein each data point is mapped on the thread level and the local lower and upper bounds are constructed simultaneously based on order statistic window.