At the end the main difference is that the sub-pixel approach is however a pixel based approach.
So sub-pixel exploits (only) the spectral information of the image; its improves the classix pixel classification because tries to identifies different spectra (and so classes) within the spectrum of the single pixel: a pixel containing only trees will have a different spectrum in respect of another one containing e.g. trees and water, it is expecetd to contain a combination of vegetation and water spectra (also) in proportion to their presence
The object-oriented approach aggregates the pixel by means of image segmentation, that is divides the image into groups of pixels (called objects) aggregating them according to criteria linked not only to the spectrum (or color), but also to the shape and homogeneity. Then for the classification the characteristics of the whole object are used, including not only the mean spectrum of all the pixel contained, but also statistical, shape, texture etc. information. It also allows to define rules based on the position of the objects in respect to each other.
Trying to compare them I would say that the object-oriented approach is more easily applicable on images which have few spectral information (e.g. it can give good results also on a single band raster like a panchromatic image) or when a more thematic approach to the classification is required (e.g. for land use, against land cover), it will also allow to combine very eterogeneous images.
On the other side the sub-pixel approach can give very good results with images having a good spectral information (e.g. >3 bands) and the indetification of the "nature" (e.g. cover) of the pixels is requierd; moreover it is more sensitive to training (e.g. capturing sample spectra of the various components): a good and complete training can give very accurate results.
The sub-pixel approach gives information about different classes within a pixel. For example, pixels where water and soil are mixed. By contrast, object oriented classification (OOC) are based on analysis supra-pixel. OOC consider the pixel spectrum, its relative spatial location, and the local spectral homogeneity and shape of adjacent cluster of similar pixels. In other words, OCC creates cluster of pixels first for creating objects (image segmentation), and then applies the classification algorithm to the objects.
Miriam
At the end the main difference is that the sub-pixel approach is however a pixel based approach.
So sub-pixel exploits (only) the spectral information of the image; its improves the classix pixel classification because tries to identifies different spectra (and so classes) within the spectrum of the single pixel: a pixel containing only trees will have a different spectrum in respect of another one containing e.g. trees and water, it is expecetd to contain a combination of vegetation and water spectra (also) in proportion to their presence
The object-oriented approach aggregates the pixel by means of image segmentation, that is divides the image into groups of pixels (called objects) aggregating them according to criteria linked not only to the spectrum (or color), but also to the shape and homogeneity. Then for the classification the characteristics of the whole object are used, including not only the mean spectrum of all the pixel contained, but also statistical, shape, texture etc. information. It also allows to define rules based on the position of the objects in respect to each other.
Trying to compare them I would say that the object-oriented approach is more easily applicable on images which have few spectral information (e.g. it can give good results also on a single band raster like a panchromatic image) or when a more thematic approach to the classification is required (e.g. for land use, against land cover), it will also allow to combine very eterogeneous images.
On the other side the sub-pixel approach can give very good results with images having a good spectral information (e.g. >3 bands) and the indetification of the "nature" (e.g. cover) of the pixels is requierd; moreover it is more sensitive to training (e.g. capturing sample spectra of the various components): a good and complete training can give very accurate results.
Sub-pixel level classification deals with performing feature the classification by breaking the pixel into more pixels based on spectral unmixing by identifying the abundance of classes using fuzzy logic, whereas, objective oriented classification is based on information from a set of similar pixels (objects). These objects are classified effectively based on spectral, radiometric properties, size, shape, texture, as well as context from a neighborhood surrounding the pixels using rule based algorithms.
- Badges
- Science topic
- Similar topics
- Geoscience
- Remote Sensing