Comparison of Common Segmentation Techniques Applied to Transmission Electron Microscopy Images
- PDF / 858,926 Bytes
- 6 Pages / 612 x 792 pts (letter) Page_size
- 74 Downloads / 175 Views
0982-KK07-04
Comparison of Common Segmentation Techniques Applied to Transmission Electron Microscopy Images Thomas E. Sadowski1,2, Christine C. Broadbridge1, and John Daponte2 1 Physics, Southern Connecticut State University, 501 Crescent Street, New Haven, CT, 06515 2 Computer Science, Southern Connecticut State University, 501 Crescent Street, New Haven, CT, 06515 ABSTRACT Nanoparticles are of interest in many applications since their decreased size may give them properties that are very different from bulk material. Often nanoparticle properties such as size (diameter) and size distribution are evaluated using transmission electron microscopy (TEM). These parameters, size and size distribution, can be more easily obtained from digitized TEM images by mapping particle signal to black and background pixel to white in a process known as thresholding then performing an algorithm known as a particle analysis. The goal of this study was to compare the ability of several popular thresholding algorithms to segment TEM images. Performance of the thresholding algorithms was evaluated through qualitative and quantitative measures. Results show that the choice of a thresholding algorithm will strongly affect the results obtained from particle analysis. INTRODUCTION Nanoparticles, particles with a diameter of 1-100 nanometers (nm), are of interest in many applications including device fabrication, quantum computing, and sensing as their diminished size may give them unique properties that differ greatly from those seen in the bulk material 1 . Application of these properties often requires an increased understanding of numerous physical characteristics of a sample of nanoparticles, especially the size (diameter) and size distribution. In the past, these parameters have often been obtained from digitized transmission electron microscopy (TEM) images by manually measuring and counting many of these nanoparticles, a task that is highly subjective 2 . A particle analysis is a recently developed computer imaging technique which counts and measures objects in the binary image3 which has proven to be a more objective alternative. Prior to particle analysis, the gray scale TEM image must be segmented into foreground and background regions using a thresholding algorithm to produce a binary image. Selecting an appropriate thresholding algorithm for image segmentation is complicated by the noisy nature of TEM images and potential contrast issues related to CCD cameras, frequently used to capture these images. The focus of this study was to compare the ability of six popular thresholding algorithms to segment TEM images of latex calibration nanoparticles. The investigated thresholding algorithms included: the Riddler-Calvard, entropy, Kittler-Illingworth, maximum likelihood, Otsu, and Tsai techniques 4 . Since the size of these latex samples are known to 90 nm ± 5%, the performance of the six thresholding algorithms were able to be both qualitatively and
quantitatively assessed. Qualitative examination consisted of comparing a gray s
Data Loading...
