Estudio de la segmentación de imágenes de microestructura anatómica de la madera mediante el análisis multiresolución

Abstract

There are many problems in the national territory such as: low forest development, commercial exploitation of timber market, lack of conservation of the environment, illegal trafficking of species and others. All of the above are directly related to the lack of knowledge for forestry development, mainly given in the characterization of a species to arrive at the correctly identification and classification. Only experts, with extensive knowledge of the microscopic anatomy From a sample of wood can make a characterization for identification. This process can be arduous, take too much time, and provide an unacceptable level of error. To improve these and other aspects, we seek to provide contributions for the consolidation of a tool that registers the characterization with computational tools, which optimizes the processes established for the identification and provides a good percentage of success. This project establishes a series of initial conditions by means of a pre-processing necessary for the study of each representative image of each sample. Within each image is sought to segment or view objects of interest; In this case these objects are given by the elements of the microstructure known as the pores. Through the use of multiresolution analysis (MRA) by the use of discrete wavelet transformer (DWT), specifically with Daubechies transformation. It is proposed to establish a segmentation model with the aim of evaluating its effectiveness and concluding about its usefulness as a tool for the digital image processing service. We used the Functional Characteristics of the receiver ( ROC) which minutely considers all the tests obtained in the whole bank of study images compared to their respective manual segmentation, with the aim of correctly classifying the cases of the positive or Negative response to automatic segmentation.