Figure 3: Classification of scenedesmus samples based on opacity, area, and morphology.
From: All-passive pixel super-resolution of time-stretch imaging
(a) Bulk metrics, i.e. opacity and area, are computed from time-stretch images restored by pixel-SR. (b) Classification is improved due to the morphology metric computed from high resolution image. The histograms in (a) and (b) shows the projections of the different clusters onto the axes of opacity, area, and morphology metric. (c) Receiver operating characteristic (ROC) curve showing the classification performance with and without pixel-SR based on the morphology metric. The inset shows the reduced overlap between cluster II and III with the pixel-SR method. (d) HR image of the cell samples selected from the corresponding clusters. (Bottom right, highlighted) The aggregates of smaller colonies are mis-classified as four-daughter colony, but is clearly distinguishable in pixel-SR time-stretch imaging. The 1,368 samples in the scatter plots were pre-screened from the 5,000 pixel-SR image frames with the brightness threshold. Interactive version is available online at http://www.eee.hku.hk/~cschan/scatter_plot/.
