Intelligent Flood Fill or: The Use of Edge Detection in Image Object Extraction
Paúl André · ePrints Soton (University of Southampton) · 2005
Content-Based Image Retrieval systems can often return poor results when attempting to match images with extraneous feature information such as complex backgrounds, shadows, or frames. As a solution, this project aimed to provide a semi-automatic object extraction tool, the ‘Intelligent Flood Fill’. This tool would also help users of graphics packages, who can find current object extraction tools inefficient. The solution implemented is based upon innovations and extensions to the floodfill technique, and is both accurate and efficient in itself, and compared to existing methods. The new algorithms presented allow extraction based on scribbles over the image which gather colour data, and/or a bounding box, outside of which the algorithm will try to revert to the last major colour change. An option for sequence extraction has also been implemented for use in VRML model creation. The architecture is flexible, so as to allow the control logic to be rewritten for another language or application as desired, and the user interface intuitive. Failures can occur when the foreground and background overlap in colour space, or when the image is low resolution or noisy. Extensions include making use of dedicated image loading libraries, and the implementation of a live wire boundary system for complex images.