Post on 12-Jan-2016
description
Image processing methods for noise reduction in the TJ-II Thomson
Scattering images
Frascati, Roma, March 26-28, 2012
Gonzalo Farias*, Sebastián Dormido-Canto, Jesús Vega, Ignacio Pastor, Matilde Santos
*School of Electrical Engineering at Pontificia Universidad Católica deValparaíso (PUCV), Valparaíso, Chile. e-mail: gonzalo.farias@ucv.cl)
Fusion Data Processing Validation and Analysis
1. Introduction• The TJ-II Thomson Scattering Diagnostic• Stray-light• Possible solutions?
2. Approaches• Problem formulation• Exhaustive detection• Connected components• Region growing
3. Results• Typical algorithm used• Example of processing• Validation
4. Conclusions and Future Works
2/29Contents
1. Introduction• The TJ-II Thomson Scattering Diagnostic• Stray-light• Possible solutions?
2. Approaches• Problem formulation• Exhaustive detection• Connected components• Region growing
3. Results• Typical algorithm used• Example of processing• Validation
4. Conclusions and Future Works
3/29Contents
TJ-II Thomson Scattering diagnostic
4/29Introduction
The TJ-II TS diagnostic
collects five different types of images
5/29Introduction
BKG STR
ECH NBI COFF
TJ-II Thomson Scattering diagnostic (noise)
6/29Introduction
stray light (noise)
TJ-II Thomson Scattering diagnostic (noise)
7/29Introduction
stray light (noise)
The TJ-II TS diagnostic
collects five different types of images (revisited)
8/29Introduction
BKG STR
ECH NBI COFF
Possible solutions?
• Apply a hardware filter: There is a Notch filter (a band-stop filter) in operation, which has a large stray-light rejection, but not all noise is eliminated.
• Apply low-pass or advanced filters (e.g. wavelets), but this action will affect to entire images. This happens with all global filters.
• Apply algorithms considering some particular characteristics of noise: localization, area, density, and in general any kind of noise feature:
1. Exhaustive detection
2. Connected components
3. Region growing
9/29Introduction
1. Introduction• The TJ-II Thomson Scattering Diagnostic• Stray-light• Possible solutions?
2. Approaches• Problem formulation• Exhaustive detection• Connected components• Region growing
3. Results• Typical algorithm used• Example of processing• Validation
4. Conclusions and Future Works
10/29Contents
Problem formulation using a toy example
11/29Approaches
Original image
Goal: Eliminate part of the image recognized as noise
noise
Exhaustive detection: how does it work?
Key idea: Use the template as sliding-window in order to find coincidences in the original image.
12/29Approaches
Original image
template
Exhaustive detection: results
Key idea: Use the template as sliding-window in order to find coincidences in the original image.
13/29Approaches
template
Original Processed
Exhaustive detection: comments
14/29Approaches
• Useful when the part of the image to look for (e.g. noise) is regular and well defined.
• There is a lot of applications where this technique has excellent results: optical character recognition, automatic number plate recognition, face and pedestrian detection, etc.
• However the technique is not suitable for irregular parts such as the stray-light of TS diagnostic.
Connected components: how does it work?
There are parts of the image where the components (pixels) are connected (no space between them). Connected pixels represent a region.
15/29Introduction
Original image
region 1
region 2
region 6
region 7
region 4
region 3
region 5
Connected components: how does it work?
Key idea: Eliminate a region (R) when some condition is satisfied.
16/29Introduction
Original image
region 1
region 2
Conditions for noise:•Position (R) is on left side
•Size(R) is >= 3 pixelsProcessed image
Connected components : comments
17/29Approaches
• Useful when the part of the image to look for (e.g. noise) is irregular and not-well defined.
• The connected components or region extraction techniques are based on the image segmentation theory.
• Very nice results on the noise reduction in the TS diagnostic (we will see later), but the predicate of connection for a pixel is too strong. Therefore, some pixels quite near, but not connected, to the region are not considered as noise in this approach.
Region growing: how does it work?
Regions are built by adding pixels. The addition is performed when the pixel meets some requirements (predicate).
18/29Approaches
Original image
region 1
region 2
region 4
region 5
region 3
Region growing: how does it work?
Key idea: Eliminate a region (R) when some condition is satisfied.
19/29Approaches
Original image
region 1
region 2
Conditions for noise:•Position (R) is on left side
•Size(R) is >= 3 pixelsProcessed image
Region growing: comments
20/29Approaches
• Useful when the part of the image to look for (e.g. noise) is irregular and not-well defined.
• The region growing is also based on the image segmentation theory.
• Similar results on the noise reduction in the TS diagnostic as the previous approach, but the regions depend on the initial seeds selected.
1. Introduction• The TJ-II Thomson Scattering Diagnostic• Stray-light• Possible solutions?
2. Approaches• Problem formulation• Exhaustive detection• Connected components• Region growing
3. Results• Typical algorithm used• Example of processing• Validation
4. Conclusions and Future Works
21/29Contents
Applying region segmentation to TS diagnostic: Algorithm of connected component approach
22/29Results
Applying region segmentation to TS diagnostic: Algorithm of region growing approach
23/29Results
Applying region segmentation to TS diagnostic: Connected components example
24/29Results
Applying region segmentation to TS diagnostic: Connected components example
25/29Results
Validation Radial profiles of the electron temperature
26/29Results
1. Introduction• The TJ-II Thomson Scattering Diagnostic• Stray-light• Possible solutions?
2. Approaches• Problem formulation• Exhaustive detection• Connected components• Region growing
3. Results• Typical algorithm used• Example of processing• Validation
4. Conclusions and Future Works
27/29Contents
Conclusions and future works
• Fusion images processing can be benefits from region segmentation methods.
• From observation of several experiments, both region segmentation methods seem to be promising in order to reduce stray-light.
• Connected components approach is quite direct, and can be implemented easily, although is not so flexible.
• Region growing is much more flexible, but selection of initial seeds is not direct.
• Validation mechanisms seem confirm visual checking.
28/29Conclusions
Image processing methods for noise reduction in the TJ-II Thomson
Scattering images
Frascati, Roma, March 26-28, 2012
Gonzalo Farias*, Sebastián Dormido-Canto, Jesús Vega, Ignacio Pastor, Matilde Santos
*School of Electrical Engineering at Pontificia Universidad Católica deValparaíso (PUCV), Valparaíso, Chile. e-mail: gonzalo.farias@ucv.cl)
Fusion Data Processing Validation and Analysis