Similarity Assessment as a Dual Process Model of Counting and Measuring

Based on recent findings from the field of human similarity perception, we propose a dual process model(DPM) of taxonomic and thematic similarity assessment which can be utilised in machine learning applications.Taxonomic reasoning is related to predicate based measures (counting) whereas thematic reasoning ismostly associated with metric distances (measuring). We suggest a procedure that combines both processesinto a single similarity kernel. For each feature dimension of the observational data, an optimal measure isselected by a Greedy algorithm: A set of possible measures is tested, and the one that leads to improved classificationperformance of the whole model is denoted. These measures are combined into a single SVM kernelby means of generalisation (converting distances into similarities) and quantisation (applying predicate basedmeasures to interval scale data). We then demonstrate how to apply our model to a classification problemof MPEG-7 features from a test set of images. Evaluation shows that the performance of the DPM kernel issuperior to those of the standard SVM kernels. This supports our theory that the DPM comes closer to humansimilarity judgment than any singular measure, and it motivates our suggestion to employ the DPM not onlyin image retrieval but also in related tasks.