A particle detection method using robust PCA

Currently, the inspection of asbestos in building materials has been done manually by human inspectors. Inspectors must count 3,000 particles to judge whether asbestos is included or not. However, counting 3,000 particles is hard jobs for inspectors and the counting result is subjective. Thus, we propose the automatic particles counting by computer. Conventional method estimated the background from an input image by a median filter and particles are detected from the difference image between input and background images. However, the method depends on the filter sizes and we can not apply the median filter to peripheral regions of the input image. Therefore, we use robust Principal Component Analysis (PCA) to estimate the background and outlier (particle) without depending on the filter sizes. In experiments, we evaluate our method by using 19 microscope images and confirmed that the proposed method detected particles with higher accuracy in comparison with the conventional method.