The goal of this project is developing an automatic image processing algorithm that enables scanning digital satellite images and recognizing artificial (man-made) objects in them (buildings clusters)
The goal of this project is developing an automatic image processing algorithm that enables scanning digital satellite images and recognizing artificial (man-made) objects in them (buildings’ clusters, e.g.).
In recent years, a need in reliable methods for automatic detection of man-made objects in satellite images has arisen. This need is caused by the following reasons:
- Significant technological advancement, which enables getting high resolution satellite images in a short time. In addition, the ability to picture any point on earth by satellites exists. This technological enhancement brings a lot of information, which must be encoded quickly and accurately
- Costs – developing a recognition algorithm reduces the need of man power
- Faster Recognition – gives us operational advantage on the enemy in a situation of war (for example: fast recognition of missile racket)
The recognition algorithm proposed here uses fractal theory in order to recognize man-made objects in satellite images. Unlike natural objects, artificial objects do not posses the fractal’s properties. Therefore, fractals are poor model for artificial objects. The recognition algorithm uses this fact to differentiate between natural and artificial objects. For the purpose of recognition, the algorithm suggests to estimate the fractal dimension in a specific image area, which is defined by a two-dimensional sliding rectangular window. The fractal dimension estimation is executed using linear regression. Regression error helps us determining whether the discussed area has fractal behavior (in other words – whether it contains artificial objects).
The detection algorithm was implemented using MATLAB version 7.0. The algorithm was examined using commercial satellite images (given by MAFAT).
- Optimal detection thresholds are dynamic – their value depends on a specific image. An adaptive mechanism that determines detection thresholds is vital to make this algorithm more reliable
- The algorithm has efficient recognition ability. However, sometimes areas recognized as artificial have some drawbacks
- Algorithm’s running time is quite high for significant satellite images (several minutes). For small images, the algorithm’s running time is about a minute long
- An empiric criterion for distinguishing between “empty” and “non-empty” satellite images is proposed. It was built relying on a 20 images sample (“empty” images are images that do not contain any artificial object). This criterion can not always be true
We are grateful to our project supervisor Mr. Dori Peleg for his help and guidance throughout this work.