Hierarchical segmentation for traditional cultural pattern based on iterative compression and clustering

How to effectively segment a given traditional cultural pattern into multiple coherent segments that are “meaningful” to the human visual perception and ensure that the segments are consistent at different resolutions is still a very challenging issue and a widely used in practice. We present a new hierarchical segmentation algorithm with contents consistent for Chinese traditional cultural pattern segmenting based on the iterative compression and clustering. Firstly, by combining d-dimensional raw features of the superpixel, we introduce a superpixel Log-Euclidean Gaussian metric (SLEGM) descriptor, which can perform the effective characterization to superpixels and then realize accurate similarity measurement for subsequent segmentation. Based on the SLEGM descriptor, we present a novel hierarchical segmentation algorithm with consistent content, where the iterative compression and clustering algorithm segments the Chinese traditional cultural pattern into a series of portions. And then the improved hierarchical clustering algorithm realizes hierarchical construction and adjustment while maintaining the containment relationship of different segmentations. Experimental on the BSDS500 benchmark indicate that our algorithm outperforms the state-of-the-art hierarchical segmentation algorithms with the best \(71.6\%\) $71.6\%$ average F-measure. Both qualitative and quantitative experimental results on our challenging Traditional Cultural Pattern (TCP) dataset verify that our algorithm is more suitable for traditional culture pattern segmenting than all competing algorithms.