University of Twente Student Theses


Gestalt theory for Remote Sensing Image Analysis

Mwakisunga, Hezron Asukile (2010) Gestalt theory for Remote Sensing Image Analysis.

[img] PDF
Abstract:The increased use of Earth observation data acquired by remote sensing technology has enhanced the capability of extracting useful information about the scene being imaged. These data inherently contain information that relates to the type and spatial contents of land cover type. From a general perspective of remote sensing, according to Gestalt principle the human eye relies on visual perception to provide much of the information about the surrounding although it is greatly limited by sensitivity to only the visible part of electromagnetic energy, viewing perspective and inability to form lasting records of what has been seen. From these limitations a continuous development of technological means, increases the ability to see and record the physical properties of the earth in different spectral channels. The techniques for deriving this information from satellite images vary depending on the information required to be observed for proper representation of a phenomenon. Recently the research has been done on the use of Gestalt principle in extraction of feature from digital images. This principle is the perceptual principle of organization that uses the statistical approach based on phenomenological observations in computing geometric structures in a digital image without any prior information. Based on this principle, many experiments that were conducted by use of grey level photography without tuning parameters performed well for general image analysis. This study concentrates on the implementation of the algorithm based on general principle of perception due to Helmholtz in extraction of linear features from multispectral images. In implementation of proposed LSD algorithm different parameters such as tolerance of level line angle, image scale and detection threshold epsilon were selected and at different threshold values were tested in order to check the performance of the proposed algorithm based on Gestalt principle to remotely sensed images. The result shows that at the range of 22.5 to 60 degrees for tolerance angle, scale value of 2 and detection threshold value of 0.0 have a greater possibility of obtaining best detection results of multispectral image. To evaluate the quality of extracted results two approaches (buffer method and visual examination) were used to check the consistence of the meaningful extracted line segment from multispectral image where the extracted dataset was comparing to the reference dataset. By buffer methods where TOP10 vector data was considered as a reference dataset the results show that the dataset containing tolerance angle of 30o equivalent to 6 numbers of orientations, scale value of 2 and detection threshold epsilon of 0.0 gives better results compared to the other datasets tested. Similar results were obtained by visual examination when the objects selected and measure their distances directly from GeoEye image were considered as a true reference dataset. The results of this study shows that the use of proposed algorithm based on Gestalt principle can also be implemented to multispectral image and give a better results provided that the implementation is done by tuning parameters that control the false detections of line segment in the digital image. Keywords: Gestalt theory, image analysis, multispectral image, perception, Helmholtz principle, large deviation, gradient orientation, parameter, landsat image, TOP10 vector data and GeoEye image
Item Type:Essay (Master)
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Programme:Geoinformation Science and Earth Observation MSc (75014)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page