Microscopia multimodal prática: registro automático de imagens de microscopia ótica e de microscopia eletrônica de varredura

• Autor(a) principal: Lima, Marcos Paulo Galdino de
• Data de Publicação: 2014
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Biblioteca Digital de Teses e Dissertações da UERJ
• Texto Completo: http://www.bdtd.uerj.br/handle/1/7698
• topic: Characterization; Iron ore. Copper Ore; Multimodal Microscopy; Image Analysis. Co-located Microscopy; Caracterização; Minério de Ferro; Minério de Cobre; Microscopia Multimodal; Análise de Imagens; Microscopia Colocalizada; Processamento de imagens; Microscopia eletrônica de varredura; Visão por computador; Minérios; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO::METODOLOGIA E TECNICAS DA COMPUTACAO

The discrimination of phases that are practically undistinguishable to the optical microscope of reflected light or to the scanning electron microscope (SEM) is one of the classical problems in ore microscopy. With the aim of solving this problem it has been recently used the technique of co-located microscopy that consists in the junction of two microscopy modalities, optical microscopy and scanning electron microscope. The aim of the technique is to provide an image of the multimodal microscopy, becoming possible the identification, in mineral samples, of phases that wouldn t be distinguished by the use of one modality only, overcoming the individual limitations of the two systems. The method of register available so far in literature to the fusion of optical microscopy and scanning electron microscope images is a hard-working procedure and extremely dependent on the operator interaction, once it involves the system calibration with a standard mesh in each routine of images acquisition. Due to this reason the current technique is not practical. This piece of work proposes a methodology in order to automate the process of images register in optical microscopy and scanning electron microscopy in a way to improve and simplify the co-located microscopy technique. The proposed method may be divided in two procedures: acquisition of transformation and register of the images with the use of this transformation. The acquisition of transformation involves, first, the pre-processing of pairs in a way of performing a crude register among the images of each pair. Then, homologous points are achieved in the optical and in SEM images. In order to this, it has been used two methods, the first one was developed based in algorithm SIFT and the second was defined from the sweeping of the highest of coefficient correlation. In the following step it is calculated the transformation. Two different approaches were used: the local weighted mean (LWM) and the weighted least squares with orthogonal polynomials (MQPPO). The LWM receives as entrance what we call pseudo-counterparts, points that are distributed in a regular way in the reference image, and that reveal, in the image to be registered, the relative local dislocation among the images. Those pseudo-counterparts may be obtained by SIFT or by the method of correlation coefficient. On the other side, the MQPPO receives a group of points with natural distribution. The analysis of the images registration obtained employed as a metric the value of correlation among the obtained images. It was noticed that with the use of the proposed variants SIFT-LWM and SIF-Correlation were obtained slightly higher results than the ones from the method with standard mesh and LWM. Thus, the proposal, besides reducing drastically the operator intervention, still enabled more exact results. On the other side, the method based in the transformation provided by the minimum square pondered with orthogonal polynomial showed lower results than to the ones produced by the method that used standard mesh