The overall objective of this thesis is to present a methodology for guiding adaptations of an RBF-based relevance feedback network, embedded in automatic content-based image retrieval (CBIR) systems, through the principle of unsupervised hierarchical clustering. The main focus of this thesis is two-fold: introducing a new member of Self-Organizing Tree Map (SOTM) family, the Directed self-organizing tree map (DSOTM) that not only provides a partial supervision on cluster generation by forcing divisions away from the query class, but also presents an objective verdict on resemblance of the input pattern as its tree structure grows; and using a base-10 Genetic Algorithm (GA) approach to accurately determine the contribution of individual feature vectors for a successful retrieval in a so-called "feature weight detection process."
The DSOTM is quite attractive in CBIR since it aims to reduce both user workload and subjectivity. Repetitive user interaction steps are replaced by a DSOTM module, which adaptively guides relevance feedback, to bridge the gap between low-level image descriptors and high-level semantics. To further reduce this gap and achieve an enhanced performance for the automatic CBIR system under study, a GA-based approach was proposed in conjunction with the DSOTM. The resulting framework is referred to as GA-based CBIR (GA-CBIR) and aims to import human subjectivity by automatically adjusting the search process to what the system evolves "to believe" is significant content within the query. In this engine, traditional GA operators work closely with the DSOTM to better attune the apparent discriminative characteristics observed in an image by a human user.