Adaptive fusion of bright-field microscopy images acquired in different focal planes

• Autor(a) principal: Catanante, Victor Augusto Alves
• Data de Publicação: 2020
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Biblioteca Digital de Teses e Dissertações da USP
• Texto Completo: https://www.teses.usp.br/teses/disponiveis/55/55134/tde-10092020-164245/
• topic: Avaliação de qualidade de imagem sem referência; Bright-field microscopy; Discrete Fourier transform; Fusão de imagens multifocais; Laplacian of gaussian; Laplaciano da gaussiana; Microscopia de campo claro; Multi-focus image fusion; No-reference image quality assessment; Transformada discreta de Fourier

Microscopy is an extremely relevant technique related to tasks that deal with micrometric order structures. Its use dates back to the 17th century and tends to evolve in parallel with the evolution of human technological knowledge. Among the various applications, the fields of biological and health sciences stand out, which involve structures normally invisible to the naked eye. There are unavoidable differences in depth between the points of the surfaces and structures which yield out-of-focus blur to images. However, high quality is necessary in order to allow precise analysis in microscopy applications. In this sense, image quality assessment and image fusion are examples of techniques that may be applied to solve the issue. Recent works on such fields show that mathematical techniques such as frequency domain analysis, multiresolution analysis and convolutional neural networks are effective to quantitatively assess the quality of images. At the same time, researchers also present many novel techniques for image fusion, either based on classical tools such as edge detection or based on state-of-the-art machine learning frameworks. The aim of this work is to develop a two-stage method, consisting of a no-reference image quality assessment and an image fusion step, to perform the fusion of bright-field light microscopy images acquired in different focal planes, and propose novel bright-field microscopy image datasets of plant leaf histological samples as a benchmark for testing both quality assessment and fusion algorithms. Frequency domain analysis and statistical methods were used to obtain a quality metric and the energy of edges extracted with the Laplacian of Gaussian filter as the fusion rule. The mean Pearsons correlation coefficient obtained for the image quality method was 0.7448, while the mean spatial frequency for the image fusion method was 0.0667.