An AdaBoost-based Approach for Coating Breakdown Detection in Metallic Surfaces Francisco Bonnin-Pascual and Alberto Ortiz Abstract— Vessel maintenance entails periodic visual inspec- tions of internal and external parts of the vessel hull in order to detect structural failures. Typically, this is done by trained surveyors at great cost. Clearly, assisting them during the inspection process by means of a fleet of robots capable of defect detection would decrease the inspection cost. In this paper, a novel algorithm for visual detection of coating breakdown is presented. The algorithm is based on an AdaBoost scheme to combine multiple weak classifiers based on Laws’ texture energy filter responses. After a number of enhancements, the method has proved successful, while the execution times remain contained. Index Terms— Coating breakdown detection, Adaptive Boost- ing, Laws’ texture energy filters, Classification, Vessel inspec- tion. I. INTRODUCTION Vessels and ships are nowadays one of the most cost effective ways to transport goods around the world. Despite the efforts to avoid maritime accidents and wreckages, these still occur, and, from time to time, have catastrophic con- sequences in environmental, human and/or economic terms. Since structural failures, like coating breakdown, corrosion or cracks are the main cause of these accidents, Classification Societies impose extensive inspection schemes in order to ensure the structural integrity of vessels. An important part of the vessel maintenance has to do with the visual inspection of the internal and external parts oh the hull. To carry out this task, the vessel has to be emptied and situated in a dockyard where high scaffoldings are installed to allow the human inspectors to access the highest parts of the vessel structure (more than 30 m high). Taking into account the huge dimensions of some vessels, this process can mean the visual assessment of more than 600,000 m 2 of steel. Besides, the surveys are on many occasions performed in hazardous environments for which the access is usually difficult and the operational conditions turn out to be sometimes extreme for human operation. Moreover, total expenses involved by the infrastructure needed for close-up inspection of the hull can reach up to one million dollars for certain sorts of vessels (e.g. Ultra Large Crude Carriers, ULCC). Therefore, it is clear that any level of automation of the inspection process that can lead to a reduction of the inspection time, a reduction of the financial costs involved and/or an increase in the safety of the operation is fully justified. This work is partially supported by FP7 project SCP8-GA-2009-233715 (MINOAS). F.Bonnin-Pascual and A. Ortiz are with the Department of Math- ematics and Computer Science, University of Balearic Islands, Spain {xisco.bonnin, alberto.ortiz}@uib.es With this aim, this paper presents a novel approach for visual detection of coating breakdown on metallic surfaces. The method is based on an Adaptive Boosting (AdaBoost) [7] scheme that chains different weak classifiers to obtain a single strong classifier. The good performance of AdaBoost has been already proved in literature. One of the most representative examples is the Viola-Jones classifier [9], [10], [6] which is able to robustly detect complex structures, e.g. faces, in real-time. In the present proposal, each weak classifier is imple- mented using different Laws’ texture energy filters [5]. Furthermore, the algorithm has been enhanced introducing texture roughness information and a colour-based filter. The rest of the paper is organized as follows: Section II describes the Laws’ texture energy filters used as weak classifiers, Section III describes how the filter responses are combined using different versions of AdaBoost to perform the coating breakdown detector, in Section IV preliminary results are shown and the performance of the algorithm is assessed, Section V presents some improvements to the algorithm and analyzes the consequent changes observed in its performance; finally, Section VI concludes the paper. II. LAWS’ TEXTURE ENERGY FILTERS Laws’ texture energy filters allow material characterization since they are able to enhance different features of its texture. For example, the following five 1D five-component basic filters can be used to detect different features: level, L5 = [1 4 6 4 1] edge, E5 = [-1 -2 0 2 1] spot, S5 = [-1 0 2 0 -1] wave, W5 = [-1 2 0 -2 1] ripple, R5 = [1 -4 6 -4 1] In this work, to describe a texture, the corresponding gray- level patch is convolved with a set of energy filters (T ⊗ filter → c) and different statistical measures are taken over an N × M neighborhood of the filter response, which finally constitute the texture descriptors: µ = ∑ N,M u=0,v=0 | c(u, v) | NM , (1) σ =     N,M  u=0,v=0 (c(u, v) - µ) 2 NM , (2) µ + = ∑ u,v|c(u,v)>0 c(u, v) NM , (3)