FUZZY AND NON-FUZZY APPROACHES FOR DIGITAL IMAGE CLASSIFICATION

摘要

This paper classifies different digital images using two types of clustering algorithms. The first type is the fuzzy clustering methods, while the second type considers the non-fuzzy methods. For the performance comparisons, we apply four clustering algorithms with two from the fuzzy type and the other two from the non-fuzzy (partitonal) clustering type. The automatic partitional clustering algorithm and the partitional k-means algorithm are chosen as the two examples of the non-fuzzy clustering techniques, while the automatic fuzzy algorithm and the fuzzy C-means clustering algorithm are taken as the examples of the fuzzy clustering techniques. The evaluation among the four algorithms are done by implementing these algorithms to three different types of image databases, based on the comparison criteria of: dataset size, cluster number, execution time and classification accuracy and k-cross validation. The experimental results demonstrate that the non-fuzzy algorithms have higher accuracies in compared to the fuzzy algorithms, especially when dealing with large data sizes and different types of images. Three types of image databases of human face images, handwritten digits and natural scenes are used for the performance evaluation.