The chief objective of this project is four-fold; (1) To develop a software module for fully automated boxing of images of single particles; (2) To use geometry-driven diffusion methodology to preprocess the particle images to denoise and enhance them, and to develop algorithms to automate the parameter selection of these schemes; (3) To use our two-stage approach to extract particle boundaries. This stage relies on contour motion estimation via Fast Matching Method and the level set algorithms; (4) To develop a toolbox of filters (criteria such as area, axial ratio, integrated intensity, perimeter-to-area ratio) that can be employed in a project-specific way to reject false hits. The main thrust of the project is a method we use for noise suppression and the subsequent contouring. We view image grids as surfaces in higher dimension and exploit their geometric properties such as curvature and edge-discontinuities to invent initial-valued partial differential equations. These governing equations are expressed as necessary conditions to minimize a specific feature-preserving energy functional and are solved using very fast Additive Operator Split (AOS) methods. We then proceed to use the noise-reduced image to detect the boundaries of the particle projections, the so-called """"""""'boxing step"""""""". Here we are exploring the accuracy of the traditional cross-correlation methods compared to that of a new scheme based on fast contour propagation that results in automatic particle boundary.
Showing the most recent 10 out of 22 publications
