Tissue-like p system for region-based and edge-based image segmentations

• Main Author: Yahya, Rafaa Ismael
• Format: Thesis
• Language: English
• Published: 2017
• Subjects: QA75 Electronic computers. Computer science
• Online Access: http://eprints.utm.my/id/eprint/81692/1/RafaaIsmaelYahyaPFC2017.pdf; http://eprints.utm.my/id/eprint/81692/; http://dms.library.utm.my:8080/vital/access/manager/Repository/vital:126042
• Institution: Universiti Teknologi Malaysia

Membrane Computing (MC), a relatively recent branch of natural computing is an emerging field in molecular computing. MC aims at abstracting models, called membrane systems or P systems, which mimic the function and structure of a biological cell. Many studies have utilized MC in various applications such as image segmentation. Due to the high computational cost of conventional segmentation techniques, bio-inspired models including MC may be applicable to tackle this limitation. In this study, tissue-like P systems, a variant of MC, with sophisticated communication rules were developed to improve regionbased and edge-based segmentation algorithms for manual and automatic segmenting of 2D artificial and real images. Manual segmentation was applied for artificial images, whereas, the automatic segmentation was applied for artificial and real medical images. The manual segmentation of 2D artificial images was achieved using four, six and eight adjacency pixel connectivity relationships, whereas, the automatic segmentation of 2D artificial and real medical images were achieved using four and eight adjacency pixel connectivity relationships. Two methods were used to realize the automatic edge-based and region-based segmentations. The first method is for 2D artificial images using P-lingua linked to Java Netbeans using the P-linguaCore4 Java Library. The second method is for the 2D real and real medical images using C# linked to P-linguaCore4 Java library. The results of the second method demonstrated the ability of the system to automatically segment 2D real and real medical images with arbitrary sizes and different image formats. The experimental results statistically proved that the methods markedly outpaced the state-of-the-art methods of 2D real image segmentation using the same data set. Furthermore, the methods showed better segmentation accuracy and ability to deal with images of different sizes and types. Extra efficient results such as reducing the number of rules and computational steps were achieved for 2D hexagonal artificial images based on Tissue-like P systems. The main contributions of this study are automatic loading and codifying of the input image as well as automatic visualization of output images after segmentation. Furthermore, six and eight adjacency pixel connectivity relationships should be considered for reducing computational steps, number of rules used and processing time in molecular computing.