Asbestos Detection from Microscope Images using Support Vector Random Field of Local Features

Y. Moricuhi, K. Hotta, and H. Takahashi (Japan)


SVRF, CRF, SVM, Asbestos


In this paper, an asbestos detection method from microscope images is proposed. Asbestos particles are characterized by needle-like shape and different colors in two specific angles of polarizing plate of microscope. Practically, human examiners use the color information and the edge information to detect asbestos. We develop the detector-based Support Vector Machine (SVM) based on local edge feature as well as local color feature [1]. The SVM detector is combined with Conditional Random Field (CRF) to correct errors to take into account the relation with neighboring pixels. To reflect the needle-like shape of asbestos particles we applied finger-shaped rectangular neighboring region instead of square neighboring region in proposed CRF.The computer experiments show that the accu racy of asbestos detection is improved by using edge features and rectangle neighboring region in addition to color features.

Important Links:

Go Back