Statistical Image Filtering and Denoising Techniques for Synthetic Aperture Radar Data Troy P. Kling Mentors: Dr. Maxim Neumann, Dr. Razi Ahmed.

Statistical Image Filtering and Denoising Techniques for

Synthetic Aperture Radar Data

Troy P. KlingMentors: Dr. Maxim Neumann, Dr. Razi

Ahmed

Outline

• Introduction– Description and example of speckle noise– Overview of local Filters (Boxcar and Lee)

• Non-Local Filters– Buades’ Non-local means filter– Deledalle’s NL-InSAR filter

• Continuing & Future Research– The new multi-baseline NL-InSAR filter– NL-PolSAR for polarimetric data– Randomized non-local means filter– Edge detection, object classification, and computer

vision

Speckle Noise

• Synthetic aperture radar (SAR) is inherently affected by speckle noise.

• Speckle can be modeled by a circular complex Gaussian distribution:

Random walk that generates a resultant complex value, i.e. multiplicative speckle noise.

Speckle Noise

Left: Google Earth image of a golf course in Harvard Forest, Massachusetts.Right: UAVSAR image of the same golf course. Speckle noise is very apparent.

Local Filters

• Boxcar filter– Local noise reduction– Moving average

• J. S. Lee’s filter– Adaptive noise reduction– Uses directional masks– – Adaptive filtering

coefficient, k, quantifies local homogeneity

Eight directional masks used in Lee’s filter.

Top left: Original image. Top right: Image with Gaussian white noise added. Bottom left: 7x7 Boxcar filter. Bottom right: 7x7 Lee filter.

Non-Local Filters – NL-Means

• Considers all pixels in the image, and performs a weighted average:

• Better at preserving textures and fine structures than most local speckle filters.

Non-Local Filters – NL-Means

Non-Local Filters – NL-InSAR

• Non-Local Means applied to interferometric SAR (InSAR) images

• Uses a more statistically-grounded similarity criterion than NL-means

• Estimates reflectivity, phase, and coherence simultaneously using weighted maximum likelihood estimation

• Applied iteratively

Non-Local Filters – NL-InSAR

Left: Google Earth image of a golf course in Harvard Forest, Massachusetts.Right: UAVSAR image of the same golf course. Speckle noise is very apparent.

Non-Local Filters – NL-InSAR

Left: Estimated reflectivity after 1 iteration. Search window of size 35x35 and similarity window of size 7x7 were used. Right: Estimated reflectivity after 10 iterations.

Non-Local Filters – NL-InSAR

Left: Estimated phase after 1 iteration. Search window of size 35x35 and similarity window of size 7x7 were used. Right: Estimated phase after 10 iterations.

Non-Local Filters – NL-InSAR

Left: Estimated coherence after 1 iteration. Search window of size 35x35 and similarity window of size 7x7 were used. Right: Estimated coherence after 10 iterations.

Continuing Research

• Multiple Baseline NL-InSAR– Extending NL-InSAR to work with more

than two SLC images

– Requires estimating the phase and coherence between several pairs of SLC images

– Similarity likelihood derivation becomes complicated very quickly

Future Research

• NL-PolSAR filter– Modifying NL-InSAR to work with polarimetry– Applications to land cover type classification

• NL-MC filter– Adding randomness (Monte Carlo methods) to

make the NL-means algorithm truly non-local

• Edge Detection– Using image filters to improve edge detection

and object classification in computer vision

References1. J. S. Lee, M. R. Grunes, G. de Grandi, "Polarimetric SAR Speckle Filtering and Its

Implications for Classification", IEEE Transactions on Geoscience and Remote Sensing, pp. 2363-2373. 1999.

2. X. X. Zhu, R. Bamler, M. Lachaise, F. Adam, Y. Shi, and M. Eineder, "Improving TanDEM-X DEMs by Non-Local InSAR Filtering", European Conference on Synthetic Aperture Radar, pp. 1125-1128. 2014. J. S. Lee, "Digital Image Enhancement and Noise Filtering by Use of Local Statistics", IEEE Transactions on Pattern Analysis and Machine Intelligence, pp. 165-168. 1980.

3. J. S. Lee, "Refined Filtering of Image Noise Using Local Statistics", Computer Graphics and Image Processing, pp. 380-389. 1981.

4. C. Deledalle, L. Denis, F. Tupin, "NL-InSAR: Non-Local Interferogram Estimation", IEEE Transactions on Geoscience and Remote Sensing, pp. 1-11. 2010.

5. A. Buades, B. Coll, and J. M. Morel, "Image Denoising Methods. A New Nonlocal Principle", Society for Industrial and Applied Mathematics, pp. 113-147. 2010.

6. C. Deledalle, L. Denis, F. Tupin, A. Reigber, and M. Jager, "NL-SAR: a unified Non-Local framework for resolution-preserving (Pol)(In)SAR denoising", pp. 1-17. 2014.

7. N. Goodman, "Statistical Analysis Based on a Certain Multivariate Complex Gaussian Distribution (an Introduction)", Annals of Mathematical Statistics, pp. 152-177. 1963.

8. "Speckle Filtering", The Polarimetric SAR Data Processing and Educational Tool, pp. 1-12. 2011.