O papel da função custo na otimização de um descritor de formas multiescala

Abstract

Parameter adjustment of shape descriptors is a challenging task in computer vision and image processing. Rather than manual schemes, which are time consuming and tedious, there is a trend towards applying metaheuristic optimization to fully support parameter settings of shape descriptors. Nevertheless, the problem of cost function selection, which plays an important role in the optimization process, has not yet been fully addressed. This work investigates the influence of the cost function on the performance of an optimized multiscale shape descriptor using three distinct clustering validation indices: the Silhouette, Davies-Bouldin and Calinski- Harabasz indices. Here, we optimize the scale parameters of the normalized multiscale bending energy descriptor using the simulated annealing metaheuristic; both classification and retrieval experiments are conducted using a synthetic shape dataset (Kimia 99), two plant leaf datasets (ShapeCN and Swedish) and the National Library of Medicine (NLM) pill image dataset (NLM Pills). The performance evaluation, in terms of the Bulls-eye and Accuracy measures, showed that optimized descriptor with the Calinski-Harabasz cost function underperformed the other functions in datasets where there is high level of dissimilarity between classes. Particularly for the NLM Pills, where each class has a well-defined pattern and furthermore the differences within pill classes are quite small. Thus, the Normalized Multiscale Bending Energy descriptor did not benefit from the optimization methodology.