Machine Vision and Applications (2016) 27:1161–1174 DOI 10.1007/s00138-016-0795-1 SPECIAL ISSUE PAPER Topology-based image segmentation using LBP pyramids Martin Cerman 1 · Ines Janusch 1 · Rocio Gonzalez-Diaz 2 · Walter G. Kropatsch 1 Received: 30 November 2015 / Revised: 12 June 2016 / Accepted: 27 June 2016 / Published online: 28 July 2016 © Springer-Verlag Berlin Heidelberg 2016 Abstract In this paper, we present a new image segmenta- tion algorithm which is based on local binary patterns (LBPs) and the combinatorial pyramid and which preserves struc- tural correctness and image topology. For this purpose, we define a codification of LBPs using graph pyramids. Since the LBP code characterizes the topological category (local max, min, slope, saddle) of the gray level landscape around the center region, we use it to obtain a “minimal” image rep- resentation in terms of the topological characterization of a given 2D grayscale image. Based on this idea, we further describe our hierarchical texture aware image segmentation algorithm and compare its segmentation output and the “min- imal” image representation. Keywords Local binary patterns · Irregular graph pyramid · Primal and dual graph · Topological characterization · Image segmentation Author was partially supported by IMUS and Spanish Ministry under grant MTM2015-67072-P (MINECO/FEDER, UE). B Ines Janusch ines@prip.tuwien.ac.at Martin Cerman mcerman@prip.tuwien.ac.at Rocio Gonzalez-Diaz rogodi@us.es Walter G. Kropatsch krw@prip.tuwien.ac.at 1 PRIP group, TU Wien, Vienna, Austria 2 Applied Math Department, School of Computer Engineering, University of Seville, Seville, Spain 1 Introduction Given a grayscale digital image I , the local binary pattern LBP( I ) [19, 21] is again a grayscale digital image which represents the texture element at each pixel in I . This is cur- rently the most frequently used texture descriptor [15] with outstanding results in applications ranging from object detec- tion [18] to segmentation [4, 11] and classification [25, 27]. Considering image segmentation, the main idea is that a good segmentation can capture perceptually important regions, which reflect local and/or global properties of the image [24]. These regions can then be used for classifica- tion and higher level tasks such as image understanding. Existing segmentation algorithms are based on threshold- ing, histograms, edge detection, split and merge strategies, watershed transformation or graph partitioning [26]. LBPs have already been used in segmentation approaches in the past: first by Ojala et al. who presented an unsu- pervised three-phase algorithm in [20]. In this paper, we now present an evaluation of our LBP-based texture aware segmentation [2] using the Berkley segmentation dataset [16]. Contrary to existing LBP-based image segmentation approaches, we do not use histograms of LBPs, but keep the spatial information of the LBPs and consider it in the seg- mentation process. Typically, the LBP operator is applied to all 3 × 3 image windows of the considered texture (region). Then the histogram provides the characteristic fea- tures of the texture. After training the feature space with the textures of interest, new textures can be classified with very good discrimination. Also for segmentation approaches based on LBPs, histograms of the LBPs were used. How- ever, due the histograms, spatial information is lost in these approaches. In [3], we proposed a new equivalent LBP encoding which transfers the code from the pixels to the neighbor rela- 123 Cerman2016a