Finding faint HI structure in and around galaxies: scraping the barrel

TL;DR

This paper demonstrates that the intensity-driven gradient filter is highly effective for detecting faint, complex HI structures around galaxies in large datasets, enabling faster and more detailed analysis.

Contribution

It introduces an optimized, adaptive filtering method for faint HI structure detection and implements it in a fast, interactive visualization environment.

Findings

The intensity-driven gradient filter enhances faint HI signals effectively.

The filter maintains high resolution of bright sources while improving faint structures.

Implementation in multi-core CPU and GPU enables real-time interactive analysis.

Abstract

Soon to be operational HI survey instruments such as APERTIF and ASKAP will produce large datasets. These surveys will provide information about the HI in and around hundreds of galaxies with a typical signal-to-noise ratio of $\sim$ 10 in the inner regions and $\sim$ 1 in the outer regions. In addition, such surveys will make it possible to probe faint HI structures, typically located in the vicinity of galaxies, such as extra-planar-gas, tails and filaments. These structures are crucial for understanding galaxy evolution, particularly when they are studied in relation to the local environment. Our aim is to find optimized kernels for the discovery of faint and morphologically complex HI structures. Therefore, using HI data from a variety of galaxies, we explore state-of-the-art filtering algorithms. We show that the intensity-driven gradient filter, due to its adaptive characteristics, is the optimal choice. In fact, this filter requires only minimal tuning of the input parameters to enhance the signal-to-noise ratio of faint components. In addition, it does not degrade the resolution of the high signal-to-noise component of a source. The filtering process must be fast and be embedded in an interactive visualization tool in order to support fast inspection of a large number of sources. To achieve such interactive exploration, we implemented a multi-core CPU (OpenMP) and a GPU (OpenGL) version of this filter in a 3D visualization environment ($\tt{SlicerAstro}$).