Development of an Extension of the Otsu Algorithm for Multidimensional Image Segmentation of Thin-Film Blood Slides

Abstract

Commencing with the Fisher discriminant, the Otsu algorithm for image segmentation is reviewed and a multidimensional extension proposed. Identification of the pixels belonging to red-blood cells in 1300times1030 24-bit colour images of thin-film microscope slides of laboratory samples of malarial infected blood is used to illustrate the Otsu algorithm and the proposed extension. The Otsu algorithm is applied to the image intensity and independently to each of the colour channels in a set of 142 images and the best combination of the colour channel outputs that maximizes the Fisher discriminant selected. ROC and MRROC curves and error rates are used to evaluate performance. The multidimensional extension of the Otsu algorithm is similarly assessed. Preliminary results obtained from sequential application of the multi-dimensional extension of the Otsu algorithm are presented and directions for further research discussed