This work describes experiments aimed at finding a straight-forward but effective way of comparing face images. We discuss properties of the basic concepts, such as the Euclidean, cosine and correlation metrics, test the simplest version of elastic templates, and compare these solutions with distances based on texture descriptors (Local Ternary Patterns). The influence of selected image processing methods (e.g. bilateral filtering) on image comparison results is also considered. Additionally, the new metric, in which differences between LTP histograms are weighted with alignment coefficients, is proposed.
展开▼
