Processing of remote-sensing (RS) imagery is characterized typically by the challenges associated with the process, such as image retrieval, storage, and analysis. The emergence of very-high-resolution (VHR) multispectral (MS) imagery and the rapid increases in data acquisition have challenged the traditional pixel-based approach to image analysis.
More recently, researchers have considered that a pixel is not a true geographical object, and it is not the optimal spatial unit for mapping a landscape. Moreover, the traditional algorithms used in pixel-based image analysis showed their inability to cope with high-resolution imagery; that is, imagery with resolution typically greater than 5 meters per pixel. These factors have contributed to the shifting of the model to, and the wider use of, object-based image analysis (OBIA).
With OBIA, a single pixel is not the subject of analysis; rather, it is a homogeneous group of pixels — image objects. An object, contrary to a pixel, provides richer information, including spectrum, texture, shape, spatial relationships, and ancillary spatial data. In turn, spatial context can be exploited to emulate a human analyst, who intuitively identifies various objects in an image, rather than individual pixels, by considering various properties, such as size, texture, shape, and the spatial arrangements of these objects to understand the semantics.
The fundamental objective of OBIA, therefore, is to use segmentation to reduce complexity of image data, and to use the calculated image objects and their corresponding attributes to develop strategies for thematic classification. With OBIA, images are segmented to make image objects from various groups of pixels to represent meaningful objects in the scene. Ontologically, this provides more accurate and reliable identification and calculation of real-world attributes from RS data, and at more appropriate scales. Moreover, it provides the opportunity to separate map data into homogenous objects of various spatial scales.
© PCI Geomatics Enterprises, Inc.®, 2026. All rights reserved.