Panchromatic IKONOS Image Classification Using Wavelet Based Features

This study investigates the use of wavelet decomposed features for panchromatic image classification for the purpose of urban land-use mapping. Discrete Wavelet Transform (DWT) is recently found to be a promising tool in image analysis of both spatial and frequency domain, as DWT has the ability to examine the signal at different resolutions and desired scales. Although DWT has been applied in different image analysis applications such as image fusion, image compression and edge detection, it is less applied in image classification technique. In this study, a Very High Resolution (VHR) IKONOS satellite image in panchromatic (PAN) mode (1-m spatial resolution) is used to examine and assess the use of the DWT for image classification of urban areas. Experimental work are conducted by comparing the image classification accuracy of the original PAN IKONOS image and the wavelet decomposed images using Haar wavelet by applying two parametric classifiers: Linear Discriminant Analysis (LDA) and Quadratic Discriminant Analysis (QDA). A preliminary investigation has been carried out to assess the effect of wavelet decomposition level towards the classification accuracy. It is found that 5% and 3% improvement of the accuracy are recorded by applying the first level wavelet decomposition using LDA and QDA, respectively. Improvement of 7% and 6% by applying second level wavelet decomposition is also found using LDA and QDA, respectively. Although the overall accuracy is only near 40%, DWT is demonstrated as a viable and promising method to improve the PAN image classification accuracy, especially when it is introduced to determine those heterogeneous land classes in urban areas.