Microscopic motion analysis: Laplacian-of-Gaussian masks for subpixel edge detection

Images of live, contracting, isolated muscle cells were recorded at high speed with a photodiode array and a digital imaging system. Forty to 160 consecutive frames were imaged with visible light and an optical microscope objective at frame rates up to 400 Hz. The frames ranged in size from 32*128 to 128*128 pixels and 8 to 26 pixel/ mu m in magnification. Homomorphic filters were applied in the Fourier-transform frequency domain to filter out high-frequency noise and remove spatial details that partially obscured edges of cell features of interest. A separable Laplacian-of-Gaussian filter was used to outline edges between regions of different refractive index. The results from this filter were also used to mask out areas of the images that had insufficient contrast to use for measurements. Edges were calculated and tracked through a contraction first with pixel accuracy, and then with subpixel accuracy by using the facet model.<>