Remote Sensing Feature Detection and Geoinformation Retrieval Via Multiscale 2D Gabor Wavelet Transform

Remote sensing involves observation, acquisition and processing of information detected and measured from airborne or spaceborne platforms. The structural information of objects such as landmarks, highways, bridges, mountains, oceans, rivers, lakes, living creatures and moving targets could be represented by edges and contours based on the high-spatial resolution remote sensing images. The geographic information system (GIS) has been widely adopted to exhibit features, patterns, and trends on the surface of the earth. The data sources from aerial photos and satellite images are typically stored as raster images that contain composite tristimulus color values of red, green and blue (RGB). It allows people to visualize, analyze, process, interpret and archive data to synthesize the boundaries, interactions and relationships of the features, patterns, and trends. Remote sensing could be either passive or active. The passive sensing system identifies natural radiation via visible light imaging or infrared photography. The active sensing system emits its own energy to scan objects so that the reflected or backscattered radiation can be detected and measured, such as Radar or Lidar (Radio or Light Detection and Ranging). To achieve better data abstraction and management with respect to spatial 2D data analysis, integration of geovisualization and geocomputation can be introduced for dynamic data exploration so that the latent interactions could be discovered. Edges are critical features of structural information. Numerous approaches have been designed for detecting edges of high spatial resolution images using gradient based algorithms. However, most methods are sensitive to noises. The 2D Gabor filter is then presented to extract and differentiate the crucial structural information via parametric optimization, where Gabor wavelet transform will be employed for edge detection and contour tracing using convolution operation of each of three primary color components of digital images with the 2D Gabor wavelet in the frequency domain. It has been shown that multiscale Gabor wavelets are suitable for segmenting remote sensing images to explore intrinsic geographical information. Numerical simulations on diversified landscape aerial images have been carried out to show the impact of the proposed approach on spatial information analysis.