engleski

TIRAMISU Methodology for Semi-Automated Interpretation of Digital Multisensor Images

In the scope of the EU FP7 TIRAMISU project a novel methodology for semi-automatic interpretation of digital multisensor images for the purpose of detection and extraction of unexploded ordnances was developed. Image interpretation is usually being executed by a professional human interpreter, which is highly reliable but simultaneously labor intensive and time consuming procedure. Experience showed that a human interpreter can’t be replaced by automatic methods of digital image processing, while some methods of digital image processing can reduce the workload of the interpreter. Methodology is based on combination of pixel and object based image analysis where lessons and rules learned on test dataset are then applied on other images of the same scene but different locations. Pixel based image analysis like transformations to various color spaces (IHS, CIELAB), Principal Component Analysis (PCA), Independent Component Analysis (ICA) and basic raster math were used. By executing these processing new datasets with different statistical properties are created. Now we have series of artificial layers which serve as input for statistical analysis and enables us better delineation of targeted objects. Object based image analysis offered additional geometrical parameters and reduced the human error in the process of manual vectorization of identified objects. Achieved results of methodology implementation on aerial images of exploded ammunition depot in Padjene (Croatia) are presented. Validation procedure was executed on randomly selected spatial subsets of 25 images and the outputs are shown in the form of confussion matrix.

color spaces, interpretation, methodology, statistical analysis,

nije evidentirano