Wavelet Based Image Similarity Measures and Applications

Clustering is usually understood as a method of splitting data, in which objects that have some similarity in the dimension of interest are close to each other, while objects that differ from each other are located further away. For example, in the case of images, this measurement is a similar distribution of colors in the corresponding areas. As a result, the images will be assigned to the same cluster in terms of appearance in terms of color perception. It is worth considering that when solving the clustering problem, one of the difficulties is not only choosing the appropriate algorithm. When using hierarchical clustering algorithms, it is necessary to build a proximity matrix on a dataset in advance. To build it, the value of the selected proximity measure is calculated on each pair of objects. It should provide some kind of numerical assessment of the similarity or, conversely, the differences between the two objects. In this paper, proximity measures based on wavelet transformations were considered. Namely, the results of clustering with partial teacher involvement on graphs constructed using various proximity measures are compared.