Historical image annotation with textures based on moments

The article discusses the problem of search and recognition of textures on single-channel images. It also describes a related problem of image tagging, i. e. attaching text labels with texture information that allows searching of text queries. The work examines digitized black-and-white historic albums as main sources of images that restricts the use of data. One possible solution is proposed in this article. It is a construction of a classifier based on the texture segmentation method with the use of moments. This classifier requires a text label per added class. Our experiments showed positive results with about 70 % of completeness (recall) and 20 % of precision. Our method has promising prospects for further research.