A Tool for Object Descriptive Feature Extraction: Application to Image Classification and Map Updating

A software tool specifically designed for the extraction of multi-criteria features from image objects is described, as well as an application example. The input data are typically multispectral imagery and a cartographic database to define the limits of the objects, though these limits can also be provided from a different source, such as an automatic image segmentation process. The output is a table with the values of all the features computed for each object, which are presented in different formats, ready to be used as input data in classification tools. Additional interpretive graphs and images can be optionally produced, useful for a better understanding and analysis of features and objects. A complete set of features can be extracted from the images describing spectral, texture and structural properties of the objects. Structural features provide information of the spatial arrangement of different elements in the analyzed object. These are related, for instance, with planting patterns of crops in agricultural parcels. Extraction of texture and structural information is based on the computation of semivariogram, Hough transform, grey level co-occurrence matrix, wavelets transform, etc. Shape features can also be extracted from the limits of the objects. The set of features proposed are applied and evaluated in a real parcel-based classification case, using decision trees combined by means of boosting techniques. The results show the usefulness and potential of the proposed features for semi-automatic land use/land cover mapping and geospatial database updating applications.