Journal of RealTime Image Processing manuscript No. (will be inserted by the editor) Lionel Lacassagne · Bertrand Zavidovique Light Speed Labeling Efficient Connected Component Labeling on RISC Architectures Received: date / Revised: date Abstract This article introduces two fast algorithms for Con- nected Component Labeling of binary images, a peculiar case of coloring. The first one, Selkow DT is pixel-based and a Selkow’s algorithm combined with the Decision Tree opti- mization technique. The second one called Light Speed La- beling is segment-based line-relative labeling and was espe- cially thought for commodity RISC architectures. An exten- sive benchmark on both structured and unstructured images substantiates that these two algorithms, the way they were designed, run faster than Wu’s algorithm claimed to be the world fastest in 2007. Also they both show greater data in- dependency hence runtime predictability. Keywords Connected Component Labeling, run length labeling, line relative labeling, Algorithm Architecture Adequation, Rosen- feld, Selkow, Real-Time implementation, transitive closure computation. Introduction Binary Connected Component Labeling (CCL) algorithms are widely used in the Image Processing field (Fig. 1). They belong to a wider class of problems in the Graph Theory area and deal with graph coloring and transitive closure compu- tation. CCL algorithms play a central part in machine vi- sion, because they often constitute a mandatory step between low-level image processing (filtering) and high-level image processing (recognition, decision). As such, CCL algorithms Lionel Lacassagne Institut d’Electronique Fondamentale (IEF/AXIS) Universit´ e Paris Sud E-mail: lionel.lacassagne@u-psud.fr Bertrand Zavidovique Institut d’Electronique Fondamentale (IEF/AXIS) Universit´ e Paris Sud E-mail: bertrand.zavidovique@u-psud.fr have a lot of applications and derivate algorithms like con- vex hull computation, hysteresis filtering or geodesic recon- struction. In its most common version, CCL is completed by two coupled finite state automaton running at the same location p respectively on the initial image (data) and the result im- age (labels). Both automatons transform a common set of neighbor pixels, the predecessors along the image scan, into the label of p depending on their value as a data and their attributed label. Due to its limited horizon, such an automa- ton artificially generates multiple labels for a given region – e.g. in cases of a concavity (Fig. 3) – to be noticed at the bottom of the concavity and resolved at the end of the image scan. CCL can address pixel sets to 1 (objects or regions out of convention) or, concurrently, both pixel sets to 1 (objects) and zero (background out of convention). 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 predecessor pixels 8 8 8 8 8 3 3 4 9 4 9 2 8 10 8 6 7 7 5 7 6 6 6 6 6 6 6 1 1 1 1 1 1 1 1 1 1 1 8 8 8 8 8 3 3 3 8 3 8 2 8 8 8 6 6 6 5 6 6 6 6 6 6 6 6 1 1 1 1 1 1 1 1 1 1 1 current pixel predecessor labels current label px p2 p4 ex e2 e4 image of pixels image of labels Fig. 1 Example of 4-connected binary components labeling Due to their importance in vision, a lot of CCL algo- rithms have been developed in the past. Some typical ones are described in the following sections. Designing a new algorithm is then a challenging task both considering the overwhelming literature and from the very performance of best existing algorithms. It is comparable to developping a