A Preprocessing Scheme for Fluorescence Microscopy Image Segmentation

Abstract

Significant technological advancements in image acquisition and enhancement techniques together with significant advancements and results in fluorescence microscopy have lead to an overload of image data, which, on any level of tractability in time and quality, is too much to be analysed manually. There is a natural demand for highly accurate automated analysis. In spite of the significant advancements, there are still problems that present themselves in a way that makes high level analysis difficult. Typically, image analysis is preceded by segmenting the regions of interest. In our case we are concerned with cells (and possibly intracellular components), bacteria, etc. This necessitates a greater degree on the quality of segmentation algorithms. This paper proposes a preprocessing scheme to draw out and enhance edge and intra-region properties to allow more accurate segmentation. This scheme is tailored towards fluorescence images but can be applied to other domains. The final segmentation results show a general boost in classification accuracy as well as a smoother fitting of the contour around the cells.