Adaptive filtering using gradient morphology
In remote sensing high resolution satellite imagery is widely used. One of the applications of remote sensing is the classification of natural resources. A classical method such as the maximum likelihood classifier, assumes that each class cover has a Gaussian distribution of the reflectance values. Discrimination takes place using all available channels of the image. To improve classification results, image pixels can be grouped into regions i.e. a partitioning of the image (segmentation) can be performed. After the grouping every region is characterised with only one reflectance value in the spectral domain. A method is presented to remove noise from images as a preprocessing step for segmentation. It is based on the analysis of the slope of the gradient profile in a moving window after sorting its elements. By classifying the slope segments, rules can be applied for smoothing. These rules can be adapted for edge preservation.
Bibliographic Reference: Paper presented: 11th IAPR International Conference on Pattern Recognition, The Hague (NL), August 30 - September 3, 1992
Availability: Available from (1) as Paper EN 37337 ORA
Record Number: 199310358 / Last updated on: 1994-11-29
Original language: en
Available languages: en