CardLin / SFEGO_OpenCL

This project is mimic of Multi-dimensional Ensemble Empirical Mode Decomposition (MEEMD) and this project achieve 10000x faster than MEEMD. Also the result is better than Bi-dimensional Empirical Mode Decomposition. (BEMD)

Spatial Frequency Extraction using Gradient-liked Operator (SFEGO)

OpenCL Version

Introduction

• In 1D signal, we can use Empirical Mode Decomposition (EMD) to analysis the signal and decomposetition the signal to several Intrinsic Mode Function (IMF) that can see the different frequency band's signal that composition the input signal.

• In 2D or higher dimension signal, there is Bi-dimensional Empirical Mode Decomposition (BEMD) and Multi-dimensional Ensemble Empirical Mode Decomposition (MEEMD) can decompose the 2D signal into different Bi-dimensional Intrinsic Mode Function (BIMF).

• Now, we are using Gradient-liked Operator that can choose different Radius=Wavelength (different spatial frequency) to do Differental on 2D signal to get the vector map (magnitude and direction=phase) Then we can do Integral on vector map to get Spatial Frame that contain such a spatial frequency information in specific Radius.

• So... this is a mimic of MEEMD. The BIMF1, BIMF2, BIMF3, BIMF4 and BIMF5 of MEEMD is very similar to R=1, R=2, R=6, R=12 and R=27 of SFEGO.

• Our work about this project is starting with name Gradient-based Multi-dimensional Empirical Mode Decomposition (GMEMD) but finally we don't use EMD... so we change the title to Spatial Frequency Extraction using Gradient-liked Operator (SFEGO)

Hardware Requirement

• Require GPU to execute OpenCL Kernel Code

• Recommend to use NVIDIA GPU with 1GB+ VRAM (VRAM usage is depend on Image Size)

• AMD Integrated GPU and Intel Integrated GPU can also run this project

• Although It can also run OpenCL on CPU mode but even the Intel Integrated GPU is faster than high-end CPU

Software Introduction

GMEMD_GPU_code_csv_LeastSquares_AmplitudeCalculation

• The grayscale analysis
• Without normalize on the result of .csv file (such amplitude is calculate by SFEGO)
• Using Least Square (SVD) on the end of SFEGO to mimic MEEMD decomposition (sum of all BIMFs should equal to input)
• Sum of all Spatial Frame is GMEMD_Combine with Least Square SVD Weight
• Sum of all Spatial Frame is GMEMD_CalcCombine

GMEMD_GPU_Color_Mulitple_Image_4K_Real16bit

• The color image analysis with 4Kx4K image
• Some Integrated GPU can't handle 4Kx4K high resolution with blank output

Execution

• Drag image file (png, jpg, bmp) and drop onto the ExecuteFile/*.exe can start execution on such image

• Command Line Execute: ExecuteFile/*.exe GalaxyCenterBlackHole.png

• If your system have more than one platform (AMD, Intel, NVIDIA is platfrom) you can choose the platfrom by changing *.exe (Platform0_GPU, Platform1_GPU, Platform2_GPU)