The main goal of this project title " “Interactive Semantic Video Search with a large thesaurus of machine-learned Audio-Visual concepts” is to make a significant contribution in the quest for semantic video search engines. More details about this project can be found here.
The main goal of DynaVis is the development of machine learning methods for machine vision systems in production and manufacturing to achieve dynamically reconfigurable systems. The project involves key players in the field of machine learning with a particular focus on machine vision. Companies from the machine vision industry and end-users from various fields complement the consortium.
Multispectral imagery in prostate cancer pathology has progressed greatly in the last 5 years. Unlike conventional RGB colour space, multispectral images allow the acquisition of a large number of spectral bands within the visible spectrum, resulting in a large feature vector size. For such a high dimensionality, pattern recognition techniques suffer from the well-known curse of dimensionality problem along with increase in execution time.
This thesis investigates novel classification algorithms for prostate cancer classification using multispectral images and the suitability of reconfigurable computing to speedup medical image classification problems. A multispectral computer vision system for automatic diagnosis of prostate cancer is implemented. Different parallel architectures for various steps in automatic diagnosis are proposed and implemented on Field Programmable Gate Arrays (FPGAs). The performance of the proposed system is assessed and compared against a microprocessor based solution. Results show that the FPGA has exactly the same classification accuracy when compared with µP based solutions but FPGA is 30 times faster than µP. Furthermore, novel algorithms based on intermediate memory tabu search are also proposed in this thesis to improve the classification accuracy. The experiments have been carried out on a number of prostate cancer textured multispectral images and results have been assessed and compared with previously reported work. The results indicate a significant improvement in classification accuracy.
Full Thesis Download: Thesis.pdf
The research and industrial communities have recently been striving to make the Internet capable of supporting unicast and multicast traffic sources with Quality of Service (QoS) guarantees. QoS parameters targeted are guaranteed throughput, end-to-end delay, and delay variation. A multicast tree for a given multicast source is a tree rooted at the source and all its leaves being members in the multicast group. Tree cost is measured by the utilization of tree links. Also tree cost is highly correlated with the number of Steiner tree nodes, i.e., nodes which are not members of the multicast group. In this thesis, a tabu search algorithm is proposed for three different multicast routing problems involving the above QoS parameters. Our proposed algorithms are then compared with other proposed techniques on numerous sample networks. On all tests, the proposed tabu search algorithms were able to find better multicast trees than those reported in the literature. Also, a greedy algorithm is proposed for multicast routing problem with dynamic membership.
Full Thesis Download: Thesis.pdf
Home | Teaching | Research | Publications | Photo Gallery