Extração de características de imagens para classificação da qualidade de couro caprino usando padrão binário local

Abstract

Many computer vision problems find solution in texture recognition methods. Tasks in different areas such as satellite image analysis, industrial inspection, medical image analysis and facial recognition exemplify applications where texture recognition techniques can be used. Local Binary Pattern (LBP) is among a list of different methods for texture feature extraction and has called attention of many academics in the last years and many different variations has been develop to address different kinds of real-world problems. In the Brazilian northeast, texture classification has special potential in the leather processing industry for leather quality classification, which is currently done manually and subjectively. In this scenario, the automation of the quality classification can help to improve the labeling process and at the same time to make process faster and objective. In this sense, this dissertation presents the results of applying the most traditional local binary pattern methods for feature extraction in caprine leather, as well as observing the feasibility for classify seven different types of quality classes. In this work, many different parameters for extraction are tested and LBP combinations are made in order to reach a better feature representation capable of solving the problem. The analysis also includes a comparison between KNN and SVM.