SAR Image Edge Detection Based On Beamlet

This research is supported by the Key Lab for Radar Signal Processing Fund under Grant No. 51431020204HK0302, the Aeronautical Science Foundation of China under Grant No. 04I53070, and the NSFC under Grant No. 60472072. The work and contribution mainly involve three following parts.· Studying filtering based edge detection algorithms and some improvement. Majority of SAR edge detection algorithms deal with images after filtering. The performance depends on both filtering and edge detection algorithms. Many filtering and edge detection algorithms have been implemented and studied.· Improving Beamlet transform. Beamlet has its advantage in extracting linear features in image. But the traditional Beamlet Transform has some problems: Firstly, it is defined on continuous function domain. When this Transform is applied to image processing, they have to use interpolation on the discrete image to get a continuous function, and then take the transform on the continuous function. That process is complicated and cost much time. Secondly, the Recursive Dyadic partitioning algorithm in the previous Beamlet Transform often lost small lines. Thirdly, previous Beamlet Transform can not really extract singularity. Therefore, a fast Beamlet Transform is presented which is faster and can really extract linear singularity. And a Tile Recursive Dyadic Partitioning algorithm is developed which can greatly relieve the short line missing problem.· Developing a Beamlet based SAR edge detection algorithm. SAR image is one of the toughest cases for edge detection due to the existence of multiplicative speckle noise which is inevitable because of interference between scatterers illuminated coherent waves. Traditional edge detectors are based on the point singularity essentially, so the conversely influence of noise still exists. We solved the edge detection problem in a completely different idea. Here we defined a linear singularity. Noise points only have point singularity but don't have linear singularity. Edges have both point singularity and linear singularity. So the linear singularity can distinguish the noise points from edge points well. According to the singularity and improved Beamlet Transform, a new edge detector which can extract edges efficiently in low SNR image is developed which can effectively extract edge and reducing the conversely impact by speckle noise.