Segmentation for the optical micrograph of fungi by using spectral transmittance and neural network

摘要

Abstract: Transmittance spectra of optical micrograph of fungi were estimated from seven band images based on the Wiener estimation method. The optical microscope adjusted with monochrome digital camera (Kodak DCS420m, 1536 $MUL 1024 pixels) with seven band filters was used for image acquisition. Transmittance spectra of 16 color transparencies were measured by spectral photometer in advance, and the seven band images of color transparency were taken by microscopic imaging system. A matrix for Wiener estimation was calculated using these measured transmittance spectra and camera responses. As the result of Wiener estimation, normalized root mean square error (NRMSE) between the original and estimated transmittance spectra of 16 color transparencies was 0.035. Wiener estimation method was applied to estimate the spectral transmittance of five species which belong to one genus of fungi. Transmittance spectra of fungi were calculated from camera responses in each pixel and the above estimation matrix. The estimated transmittance spectra were used for segmentation of conidia and hyphae in fungal microscopic image to identify them. Competitive learning in neural network was applied to the segmentation from spectral microscopic image. It was found that segmentation based on spectral transmittance was more accurate than the segmentation based on RGB values. !20