Fatema A. Albalooshi, Alex Mathew
Download Project (640 KB)
Image segmentation is a very mature field that is used in several applications such as medical imaging, machine vision, object detection, object recognition, traffic control systems, and many more. Several general-purpose algorithms and techniques have been developed for image segmentation and fast implementations and libraries are available. Water body segmentation in aerial imagery is a harder problem as the properties of water, such as reflectivity varies with several environmental factors. For instance, surface brightness changes with incident light according to time of the day, haze and cloud, angle of capture, and specular reflectivity dictated by Fresnel equations. In addition, the color of water can vary depending on the presence of micro-organisms and size of water body area. Over the past decade, a significant amount of research has been conducted to extract the water body information from various satellite images. The objective of this research is to segment out water bodies to narrow down the search regions for oil leak detection. Color, texture and gradient features are used to extract water body region. The histogram of hue, saturation, and value,( H , S and V) are concatenated together to form a 'color feature vector'. These features are used to train a Support Vector Machine(SVM) classifier. Each pixel is then classified as water or non-water based on the histogram of pixels in a 3 x 3 neighborhood around it. The location of camera, time of capture, presence or absence of sunlight, and depth of water body are challenges that have been analyzed and discussed. We have also given a comparison with other well known segmentation methods such as K means clustering, mean shift clustering, and graph cut. Important factors to be taken into consideration for future research work are also identified and discussed.
Vijayan K. Asari
Primary Advisor's Department
Electrical and Computer Engineering
Stander Symposium poster
"Water Body Segmentation in Aerial Imagery" (2013). Stander Symposium Projects. 303.