Object-based or object-oriented classification uses both spectral and spatial information for classification. The process involves categorization of pixels based on their spectral characteristics, shape, texture and spatial relationship with the surrounding pixels. Object-based classification methods were developed relatively recently compared to traditional pixel based classification techniques. While pixel based classification is based solely on the spectral information in each pixel, object-based classification is based on information from a set of similar pixels called objects or image objects. Image objects or features are groups of pixels that are similar to one another based on the spectral properties (i.e., color), size, shape, and texture, as well as context from a neighborhood surrounding the pixels. Object-based classification is a two step process, first the image is segmented or broken into discrete objects or features with and then each object is classified. This type of classification attempts to mimic the type of analysis done by humans during visual interpretation.
Image segmentation is a key component to object-based classification. Segmentation is a process by which pixels in an image are grouped into segments, objects or features, that have similar spectral and spatial characteristics. Each of these objects or features contain multiple pixels. The segments in the image ideally correspond to real-world features, for example buildings or tree crowns. There are a variety of different parameters that are used in segmentation. The scale of the objects or features is one of the important variables in the image segmentation process. The scale sets the minimum number of pixels that must be contained in a group for it to be a separate segment or object.
After an image has been segmented into appropriate image objects, the image is classified by assigning each object to a class based on features and criteria set by the user. These criteria can be broken into two general groups, characteristics related to each object (i.e. color texture) and characteristics that describe the relationship between objects. For example, we know that roads are elongated, some buildings approximate a rectangular shape, and trees are highly textured compared to grass. To classification of roads could be improved by also including the connectivity to other features identified as road as a classification criteria.
- Color or Spectral Properties: mean or standard deviation of each band, mean brightness, band ratios
- Size: area, length to width ratio, relative border length
- Shape: roundness, asymmetry, rectangular fit
- Texture: smoothness, local homogeneity
- Class Level: relation to neighbors, relation to other objects
Relationship to Other Objects
- Connectivity of Other Objects: If a specific object is connected (touching) an object of another specific class
- Proximity to Other Objects: Distance to other specific classes is considered
Object-based classification methods work well with high-resolution black and white or multispectral imagery, although they can be used with lower spatial resolution imagery as well. Object-based classification methods provides a relatively quick, automated method for identifying and extracting features like rooftops or tree crowns saving an analyst from digitizing them by hand.
Explore: Feature Extraction in ENVI
The Feature Extraction Tool performs object-based classification by segmenting an input image into objects, then classifies those objects into categories using the spatial, spectral, and/or texture attributes that it calculates for each object. ENVI offers a tutorial on using the Feature Extraction Tool.