Counting leaves without ``finger-counting'' by supervised multiscale frequency analysis of depth images from top view

Abstract

Depth imaging is applied to characterize the shoot of seedlings from top-view. We demonstrate how quantitative informations of biological interest, such as leaves counting can be extracted from such images without performing 3D reconstruction of the shoot. This is obtained from 2D Fourier multiscale analysis without any requirement to segment nor detect leaves one by one numerically. We discuss the robustness and limitations of this approach and present possible extension with 3D Fourier analysis applied to estimate the plant plastochrone or 3D + T Fourier analysis in the estimation of circadian rythms.

Watching plants grow – a position paper on computer vision and Arabidopsis thaliana

“…They over-segment the image and merge these segments using angular and depth information from the stereo disparity map obtained by block matching. Some work has been done using a depth camera to extract 3D information ( [72], not Arabidopsis), this being a relatively low cost approach. The MSU-PID database includes depth camera images of Arabidopsis [39].…”

Aprendizaje automatizado basado en transformadas wavelet para la segmentación de hojas en imágenes infrarrojas

“…Sin embargo, aplicar un algoritmo de visión en este campo no es algo trivial debido a dificultades como: (1) fondos de imagen no uniformes, (2) solapamiento de órganos, (3) similitud entre órgano y fondo y (4) cambios en la posición de los órganos respecto a la cámara. Rousseau et al [1] utilizaron el análisis multiescala de Fourier 2D sin ningún requisito para segmentar y detectar y contar hojas una a una. Aplicaron un algoritmo de clasificación supervisada capaz de contar las hojas desde 2 hasta 6 en las primeras etapas de crecimiento de la plántula desde una posición cenital.…”