Automatic classification of cells and intercellular spaces of apple tissue

A procedure was developed for the automatic identification of cells and intercellular spaces of apple tissue from microscopic images. Two cultivars were used to develop the method: ‘Golden Delicious’ and ‘Champion’. Images of tissue microstructure were obtained using a confocal scanning laser microscope (CSLM). The area, perimeter, elongation, convexity, circularity, rectangularity, shape roughness and angularity of objects were automatically segmented from CSLM images and then analyzed. These geometric features were used to classify objects into two groups by means of linear discriminant analysis (LDA). Expert-based classification was used as the reference data for the LDA model. Using “step forward” analysis and the elimination of redundant variables, four geometric parameters: area, perimeter, circularity and shape roughness, were found to be the most important for the classification of cells and intercellular spaces.

► Cells and intercellular spaces can be characterized by their size and shape.
► Classification by means of LDA enables to process a large number of images.
► Examined varieties of apples require individual classification functions.