Multi-image morphing: Summarizing visual information from similar ancient coin image regions

The process of synthetically producing an image illustrating merged parts of multiple source images is usually known as image morphing. In this work a system is presented which morphs more than two source images to one output image. Its focus lies on using ancient coin images belonging to a common coin type. Nowadays, these coins can be worn or damaged. The goal of the presented morphing framework is the automatic detection and summarization of visual data of common regions by which outliers like wear marks of coins are removed. Since image registration forms the basis of the morphing system, SIFT flow functionalities are used. The selection of possible region-candidates is inferred by means of a Markov Random Field in order to find the best combination of visual content. Finally, solving the Poisson equation smooths the morphed image. An evaluation is carried out in which the system is applied to three different data sets in order to demonstrate visually aesthetic results. A second evaluation is done by investigating a classification task of ancient coin images. It is shown that substituting a morphed image as training image in the classification task improves the representation of a coin type compared to a single image.