An efficient casts recognition algorithm in urinary sediment images

Abstract

A new and efficient method for casts recognition in urinary sediment microscopic images is proposed in this paper. It combines the shape and texture characteristics of casts, and accordingly, consists of two steps. In the first step, the casts' tube-like shape feature is expressed by a modified method stems from the traditional one which is based on the minimum bounding rectangle(MBR). Instead of using MBR, we make use of the centerline to describe its shape while curved casts are concerned. Then, in the next step, some texture features are extracted and send to the SVM classifier for further judgment. As this method is quite focus on the features of casts, both shape and texture, it is very effective to recognize casts in urinary sediment images. Large experiments proved that this method is easy to implement and achieves high accuracy.