Identification of orientation of galaxies in the Galaxy Zoo dataset using spectral clustering

TL;DR

This paper presents a spectral clustering approach to determine galaxy orientations in the Galaxy Zoo dataset, highlighting its effectiveness on clean images and limitations with noisy or complex images.

Contribution

It introduces a spectral clustering method for classifying galaxy orientations, improving understanding of galaxy morphology analysis.

Findings

Spectral clustering effectively classifies galaxy orientations in clean images.

The method struggles with noisy images and multiple galaxies in a single image.

Clustering works better on smaller, less complex subsets of data.

Abstract

This work identifies the orientation of galaxies in the Galaxy zoo data set. The images are first identified by the number of principal components required to represent 99 percent of the variance of the image. K means clustering is used to separate the galaxies on the basis of their central brightness along with outlier separation. Spectral clustering is then used to separate circularly symmetric galaxies and the remaining galaxies are identified according to their orientation as flat , left and right on the basis of the alignment of the main axis. It is also seen that spectral clustering fails to make this classification when the galaxy images are noisy and works only when applied on a smaller subset of the total number of images in the Galaxy zoo data set. This method also fails in the presence of multiple galaxies in the image, considering them as an individual entity.