Short URL

Object-Based Image Analysis

 

Other Remote Sensing areas:

Loading, please wait...

Why Object-Based?


Typical remote sensor data processing involves both raster and vector data models, with the richest suite of GIS modeling applications (e.g. watershed analysis using a LIDAR-derived digital elevation model, land cover extraction using per-pixel pattern recognition) built on the raster data model.  In recent years, exciting new techniques in remote sensor data processing have emerged utilizing object oriented constructs.  At the heart of object oriented image analysis (led by Trimble eCognition) is the concept of automated image segmentation.  An image object may be a pixel, or it may be a spatially aggregate image segment that includes two or more adjacent pixels.  Each image object has properties such as area, perimeter, shape, average green value.

There are literally hundreds of properties that may be computed for image objects and the number generally increases throughout the image processing cycle as expert information is encoded by the developer or analyst.  Applications of object oriented image analysis range from land cover maps that require varying spatial scales of analysis in different parts of the image to automated extraction of occluded road segments (hidden by tree canopy or shadow in detailed aerial imagery).   The figure below shows image segments (or objects) derived from Landsat 5 imagery over an agricultural area in the Arkansas delta.

Research at CAST:  We are interested in the automated application of object oriented image analysis.  There are two dimensions to CAST's focus, including 1) the automated development of rules for working with image objects and their properties; and 2) the application of relatively simple rules developed for a small geographic area to much larger images (e.g. consider an IKONOS image of Jamaica at 150 GB in size).  Machine learning and high throughput computing (HTC) techniques play an important role in this research area.

Scale Management:  CAST uses the infrequent term scale management when describing intelligent handling of image objects according to their spatial scale.  Spatial aggregation scale is not only a central concept in image segmentation itself but is embedded in environmental patterns (e.g. how an ecosystem varies across scales of spatial aggregation).  In general terms, as technology better manages spatial scale, then the patterns associated with scale (e.g. that are observed using remote sensing) can better inform the application of technology (e.g. determining what kinds of remote sensors and sensor configurations are really needed).