Image segmentation for depth region identification

• Main Author: Raghavan, Arvind
• Other Authors: Martin Constable
• Format: Final Year Project
• Language: English
• Published: 2013
• Subjects: DRNTU::Visual arts and music::Technique and composition; DRNTU::Engineering::Mathematics and analysis::Simulations
• Online Access: http://hdl.handle.net/10356/51056
• Institution: Nanyang Technological University

Several recent applications require landscape photographs to be divided into depth- regions (foreground, middleground, background, sky). Currently, this segmentation is done manually, since depth-perception is itself a very subjective area, and none of the recent advances in image segmentation cater specifically to the unique features of depth-planes in landscapes. A method to automate or semi-automate this segmentation would make these applications considerably less cumbersome, especially in areas where large volumes of images are to be analysed. This project explores the theory behind depth perception, the recent research done on landscape photographs and the currently dominant segmentation approaches, and proposes a method to effect depth-segmentation in landscape photographs. The proposed method isolates specific image attributes of interest to landscapes, such as inter-region contrast, intra-region contrast, lightness, saturation and hue, to arrive at effective parameters to identify these depth planes. This project also explores techniques that could enhance the segmentation results of the proposed method and presents the performance analysis from the evaluation of those techniques.