Page 323 - NGTU_paper_withoutVideo

Modern Geomatics Technologies and Applications

1- BACKGROUND

The reliability of thematic maps resulting from satellite image classification depends on how analysis of remotely sensing dates.
The results of using spectral data of pixels alone are not satisfactory. Experiments have shown that addition of the texture and
contextual information can increase the accuracy of classification. [2, 3] Object-oriented approach is mainly based on the use of
Important object characteristic such as the shape for classification. Object oriented approach to image analysis requires complete

segmentation of an image to obtain image object primitives as basic processing units.[4] The technique used for image
segmentation in eCognition software called Multiresolution segmentation. The basic aim of the segmentation process should be
to generate meaningful objects. This means that a corresponding image object represents the shape of each real-world-object.
To get segments suited for the desired classification, the segmentation process can be manipulated by defining which of the
loaded channel are to be used by what weight and by three parameters: Scale parameter, color and shape. [4, 6] This research
aims to evaluate the effects of Multi-resolution Segmentation parameters in eCognition software on the accuracy of Object-
Oriented Classification of satellite images for extracting different land use\cover.

2- STUDY AREA AND USED IMAGERY

The steady area is part of Tehran metropolitan area in Iran located between 51°21'0"E, 35°40'0"N to 51°24'0"E, 35°44'0"N.

three images used in this study are those sensor PAN(taken 08 ,sep,2003) LISS (taken 13. Oct. 2003) of IRS- ID satellite
and PAN IKONOS satellite (taken 13. Jan.2001).
Image enhancement and geometric correction are done by PCI Geomatica software, and ERDAS Imagine does pixel based

classification and accuracy assessment. All object-oriented image analysis procedures, including multiresolution
segmentation, sampling and preparing knowledge bases, and nearest neighbour and fuzzy classification, are performed
using eCognition software.

3- Pixel based classification:

According to the spectral reflection nature of urban areas and heterogeneity characterizes and different compound of
features and spectral and spatial resolution limitations of the images, the number of information classes is hardly recognized.
In this present study, four classes, such as buildings, roads, vegetation, and bare areas, are recognized. In addition to these

classes, shadow class is recognized from IKONOS image. Pixel based classification is perform by MLC algorithm. The
accuracy assessment of classification results is obtained from 30 random points for each class in both classification
methods.

4- Object-Oriented Image analysis:

The most evident difference between pixel-based image analysis and object-oriented image analysis is that first, in object-
oriented image analysis, the basic processing units are image objects or segments, not single pixels. Second, the classifiers
in object-oriented image analysis are soft classifiers that are based on fuzzy logic. Soft classifiers use membership to

2