PEL: A Predictive Edge Linking algorithm q Cuneyt Akinlar ⇑ , Edward Chome Anadolu University, Computer Engineering Department, Eskisehir, Turkey article info Article history: Received 15 December 2015 Accepted 28 January 2016 Available online 6 February 2016 Keywords: Edge detection Edge linking Edge segment detection Canny CannySR Edge Drawing (ED) Contour detection gPb abstract We propose an edge linking algorithm that takes as input a binary edge map generated by a traditional edge detection algorithm and converts it to a set of edge segments; filling in one pixel gaps in the edge map, cleaning up noisy edge pixel formations and thinning multi-pixel wide edge segments in the pro- cess. The proposed edge linking algorithm walks over the edge map based on the predictions generated from its past movements; thus the name Predictive Edge Linking (PEL). We evaluate the performance of PEL both qualitatively using visual experiments and quantitatively within the precision-recall framework of the Berkeley Segmentation Dataset and Benchmark (BSDS 300). Both visual experiments and quanti- tative evaluation results show that PEL greatly improves the modal quality of binary edge maps produced by traditional edge detectors, and takes a very small amount of time to execute making it suitable for real-time image processing and computer vision applications. Ó 2016 Elsevier Inc. All rights reserved. 1. Introduction Edge detection is a very important and fundamental first step in many computer vision and image processing applications. A tradi- tional edge detection algorithm [1–4] takes a grayscale image as input and produces a binary edge map (BEM) as output, where an edge pixel (edgel) is marked (e.g., its value in the edge map is 255), and a non-edge pixel is unmarked (e.g., its value in the edge map is 0). The binary edge maps produced by traditional edge detectors are usually of low quality, consisting of gaps between the edgels, unattended edgels and noisy notch-like structures, ragged and multi-pixel wide edgel formations, etc. An example of such an edge map with low quality artifacts is shown in Fig. 1 for the famous Lena image. This edge map was obtained by the OpenCV imple- mentation of the widely-used Canny [2] edge detector (cvCanny) [5], which is the fastest known Canny implementation. To obtain this edge map, the input image was first smoothed by a Gaussian kernel with r = 1.5 (using cvSmooth from OpenCV), and cvCanny was called with low and high threshold values set to 20 and 40 respectively, and the Sobel kernel aperture size set to 3. Fig. 1 also shows the close-up views of two separate sections of the edge map to illustrate the low quality artifacts, which can be grouped in three categories as follows: (1) There are discontinuities and gaps between edgel groups as can clearly be seen in the close-up views of the two enlarged sections of the edge map. Some of these gaps need to be filled up. (2) There are noisy, unattended edgel forma- tions and notch-like structures. This is more evident in the close-up view of the upper-left corner of the edge map. These noisy artifacts needs to be removed. (3) There are multi-pixel wide edgel formations in a staircase pattern especially around the diagonal edgel formations (both 45 degree and 135 degree diagonals). Such formations can be seen in many places in the edge map, and they need to be thinned down to 1-pixel wide chains. To improve the modal quality of the binary edge maps produced by traditional edge detectors, edge linking methods have been pro- posed in the literature [7–24]. The goals of these methods are com- monly to remove noisy edgel formations and clean up the edge map, and to fill in gaps between edgels to form longer edgel groups. Snyder et al. [7] is one of the first researchers to present a method to deal with errors in edge detector results. The authors propose a method to close gaps in edge maps while preserving edge structure and connectedness based on the concept of a cham- fer map. Xie [6] presents a method to link edge pixels for the pur- pose of line segment detection. The method makes use of the concepts of horizontal edge element and causal neighborhood win- dow to realize edge linking and consequently line segment detec- tion. Given a binary edge map, the author performs linking of pixels aligned on the same image line into what is called a horizon- tal edge element, which is then converted to a line segment. The http://dx.doi.org/10.1016/j.jvcir.2016.01.017 1047-3203/Ó 2016 Elsevier Inc. All rights reserved. q This paper has been recommended for acceptance by K. Chung. ⇑ Corresponding author. E-mail addresses: cakinlar@anadolu.edu.tr (C. Akinlar), edwardchome@anadolu. edu.tr (E. Chome). J. Vis. Commun. Image R. 36 (2016) 159–171 Contents lists available at ScienceDirect J. Vis. Commun. Image R. journal homepage: www.elsevier.com/locate/jvci