Fig. 2

Atomic column detection in the presence of noise and low contrast. The accuracy of atom detection is analyzed as a function of noise level, λ. The noise, sampled from a Poisson distribution, is added to the STEM simulated images of bulk SrTiO₃ a and SrTiO₃/BaTiO₃ (b). The accuracy is computed by comparison of detected features Kp(x,y) at some \(\lambda \ne 0\) to the ideal images (λ = 0). To demonstrate the dependency of atom detection on the contrast of the atomic column, the accuracy of the detection of O columns and “Sr + Ti” is calculated and shown separately in (a). The images indicate the detected atomic columns at the corresponding noise levels. For ease of comparison with experimental images, the noise level shown here is defined as a fraction of the highest scattered intensity (from an atomic column) that is present in the simulated image. Due to the projective nature of STEM imaging, Ti columns are in fact not pure Ti columns but mixed Ti and O columns