Classification is one of the important methods in extraction of information from digital satellite images. The traditional methods of classification are based on the value of individual pixels in the images which are reflected from territorial features. The ability of pixel based approach in the satellite image classification is limited, when objects have similar spectral information. This circumstance reduces the classification accuracy. Then in this approach the image cannot be classified correctly. The classic pixel-based approach is based on “binary theory”. By this theory, one pixel will be labeled to a class or is not assigned or remains unknown or not classified. In the case of the pixels in the overlapping areas of the feature space, by binary theory, those pixels will be labeled into only one class but they show the affinity with more than one class. With binary theory the classification result will not be accurate.
But object oriented image analysis approach is the procedure in image analysis that combines spectral and spatial information. This approach segments the pixels into objects according to the tone of the image and classifies image by treating each object as a whole. Utilizing the texture and contexture information of the object in addition to using spectral information, object orient image analysis has more powerful image analysis ability. The basic theory of object oriented approach is the fuzzy theory, in the case of the overlapping area in the feature space, pixels in the overlapping areas will not be classified only into one information class, which is not correct in the real world, but are given different membership to one (with the value 1) or more than one (with the value between 0 to 1) information classes. This approach of classification is soft classifier (for example fuzzy system), which uses a degree of membership to express an object’s assignment to a class. The membership value usually lies between 1.0 and 0.0, where 1.0 expresses full membership (a complete assignment) to a class and 0.0 expresses absolutely non-membership. The degree of membership depends on the degree to which the objects fulfill the class-describing conditions. The main advantage of this soft classifier lies in their possibility to express uncertainties about the classes’ descriptions. It makes it also possible to express each object’s membership in more than just one class or the probability of belonging to other classes, but with different degrees of membership. This classification can be done by the algorithm of nearest neighbor. The nearest neighbor is applied to selected objected features and is trained by sample image objects. The fuzzy realization of the nearest neighbor approach which is used in eCognition software automatically generates multidimensional membership functions. They are suitable for covering relations in multi-dimensional feature space. The nearest neighbor classifies image objects in a given feature space and with given samples for the class of concern.
Materials and methods
In this study the maximum likelihood classification (MLC) of pixels based and nearest neighborhood of object oriented (O.O) for classifications of satellite images are compared. This comparison is done by extracting the land cover of west Azarbaijan province. To compare these methods we used satellite images of SPOT 5 to extract land use maps of the case study area. To do so, in pre-processing stage of images, geometric correction including georeferencing, orthorectification and atmospheric correction were implemented. In processing stage, images after enhancement were classification in two ways. Frits, pixel-based classification was done based on Maximum likelihood algorithm, then object oriented classification was implemented by using the nearest neighbor algorithm in eCognition software.
Results and discussion
After satellite images classified by two methods, to evaluate and compare the results, overall accuracy and Kappa coefficient of the frame were extracted for each algorithm and it was determined that in pixel-based classification algorithm , the maximum Likelihood approach with overall accuracy of 88.37% and Kappa coefficient of 0.87 has lower accuracy in comparison with nearest neighbor algorithm, because Kappa coefficient of classification in nearest neighbor algorithm in object oriented method estimated about 0.94 while overall accuracy was about 95.10%. This means that “O.O” approach has almost 7% improvement in the overall accuracy and the Kappa indices. In another word, the object oriented image analysis can be the best method in classification of satellite images compared to pixel-based algorithms.
This research has been done to compare pixel-based algorithms and object oriented image analysis in classification of digital satellite images. The results of this research showed classification based (O.O) method provides more precise results in satellite image processing. Also it will be better to consider geodatabase and calculating geometric characteristics of each land use class in the pos-processing stage. The outcome of the research has been applied in the land cover extraction of East Azarbaijan province and extracting the land use changes of Satarkhan dam basin for the period of 30 years.