Image Feature Extraction and Similarity Evaluation Using Kernels for Higher-Order Local Autocorrelation

摘要

The Higher-Order Moment (HOM) kernel is known to enable efficient utilization of higher-order autocorrelation (HOA) features in signals and images. Several authors report that kernel-based classification methods employing this kernel can classify image textures utilizing the HOA features efficiently. This work evaluates the nature of the HOM kernel of various orders as measures for image similarity. Through sensitivity evaluation and texture classification experiments, it was found that the Local Higher Order Moment (LHOM) kernel enables to control the selectivity of the similarity evaluation by using the Gaussian window.