PROCESSING A MONOCHROMATIC EARTH OBSERVATION IMAGE FOR BUILDINGS DETECTION
The purpose of this work is to develop a new technique of image processing, in particular for buildings detection on monochromatic Earth-observation images of different types. The method is based on the property of most man-made objects which consists in straight edges and mostly right angles. Repr...
| Main Author: | |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Akademperiodyka
2017-12-01
|
| Series: | Радиофизика и электроника |
| Subjects: | |
| Online Access: | http://re-journal.org.ua/sites/default/files/file_attach/2017-4/4.pdf |
| Summary: | The purpose of this work is to develop a new technique of image processing, in particular for buildings detection on monochromatic Earth-observation images of different types.
The method is based on the property of most man-made objects which consists in straight edges and mostly right angles. Representing an image separately as gradient value- and direction-layers allows to detect targets in both cases: a bright object on a dark background and a dark one on a light background. The developed 2D adaptive image filter allows to detect straight edges even if the given image fragment has a low contrast and is extremely noised. The next processing of line-segment list without image raster works faster and allows to detect a small set of possible targets.
The developed algorithm forms a limited set of objects which shape is close to anthropogenic by processing the original image.
The proposed method can be used as an image segmentation algorithm for formation of hypotheses and subsequent recognition using neural networks. The method is also useful as a means for automating and improving the efficiency of training dataset images preparing by an operator for a neural network, and thereby as optimizing this process. |
|---|---|
| ISSN: | 1028-821X 2415-3400 |