Edge Detection and Image Filter algorithms for Spectroscopic Analysis with Deep Learning Applications

TL;DR

This paper explores the application of various edge detection algorithms to ARPES data analysis, demonstrating that the Canny filter outperforms others in noisy conditions typical of spectroscopic measurements.

Contribution

It is the first systematic application of edge detection techniques to ARPES data, identifying the Canny filter as the most effective method for noisy spectroscopic images.

Findings

Canny filter is most effective for noisy ARPES data

Other edge detection methods fail to accurately detect ARPES bands

Systematic comparison of multiple edge detection algorithms for spectroscopic analysis

Abstract

Edge detection and image filters are commonly used in computer vision. However, they have never been applied to the data analysis of angle-resolved photoemission spectroscopy (ARPES) data before in a systematic fashion. In this paper we will use the Sobel, Laplacian of a gaussian (LoG), Canny, Prewitt, Roberts, and fuzzy logic methods for edge detection in the ARPES results of HfP2, ZrSiS, and Hf2Te2P2. We find that the Canny filter is the best method for edge detection of noisy data that is typical of ARPES measurements, while the other edge detection techniques are not able to correctly detect ARPES bands.