Extracting Painted Pottery Pattern Information Based on Deep Learning

Abstract

This paper proposes a method that can effectively recover pattern information from painted pottery. The first step is to create an image of the pottery using hyperspectral imaging techniques. The Minimum Noise Fraction transform (MNF) is then used to reduce the dimensionality of the hyperspectral image to obtain the principal component image. Next, we propose a pattern extraction method based on deep learning, the topic of this paper, to further enhance the process resulting in more complete pattern information. Lastly, the pattern information image is fused with a true colour image using the improved sparse representation and detail injection fusion method to obtain an image that includes both the pattern and colour information of the painted pottery. The experimental results we observed confirm this process effectively extracts the pattern information from painted pottery.