Automatic Extraction Of Main Roads In SAR Image

Synthetic Aperture Radar (SAR) is an active microwave remote sensing imaging system, which is widely used in earth observation and military detection due to its quality of working all day, all weather and its penetration ability in certain frequency band. Developing efficient methods of SAR image processing so as to better interpret SAR images and obtain more information helps a lot in widely utilization of SAR images. Roads are arteries of transportation, linking different regions. Therefore, automatically extracting main roads in SAR images makes important sense in both military and economic application of SAR.The most difficult thing in automatic extraction of roads in SAR image is that we have to face various disturbance, including not only those beside roads such as automobiles,trees and street lamps, etc., but also speckle which floods in the whole SAR image. Speckle makes things difficult in interpreting SAR images. Speckle reduction methods always blur linear features, they do no good for road extraction; according to our problem, combining speckle reduction and road detection together would be a more reasonable choice. Both Ratio of Average (ROA) and Cross-Correlation method use local statistics to judge road. They perform good detection while smoothing speckle. Moreover, they both have Constant False Alarm Rate (CFAR) which is important in radar target detecting.There are a huge number of road candidates after detecting. The delineation of road candidate is redundant because candidates in a neighborhood have similar attribute and are strongly correlated. Phase Grouping Method (PGM) groups candidates in a neighborhood sharing similar attribute together. Therefore, we use PGM grouping road candidates to line segments and delineating segments instead of candidates. A new approach is proposed to cut off the huge computation of PGM by making a prejudgment using direction attribute of candidates.Among segments obtained by PGM, some belong to the real roads, some are false responses, there are still many parts of the roads remain undetected. In order to connect segments that belong to real roads and suppress those who are not in a road, some contextual knowledge about the structure of roads must be...