Extracting unknown repeated pattern in tiled images

Abstract

Humans can easily recognize specific patterns and their repetition in an image but it is very difficult for machines to do. However, the machines can create a plethora of repeated patterns and tiled images. This research paper proposes some methods of recognizing an unknown repeated motif in tiled images in computer generated raster graphics. Three approaches, autocorrelation of an image, comparing with template strip and cyclic bitwise XOR-ing of the image are compared in this paper. Finally, the third algorithm is proposed as it detects locally repeating unknown motifs in a tiled image and outperforms the former two methods in robustness and provides a reliable result. Unlike other traditional approaches, the proposed method does not require any feature extraction and clustering of features or patches and it is unsupervised.