Section of Life Sciences and Technologies, EPFL
In 2020, ESA Euclid space telescope will be launched in orbit to capture images of galaxies from the entire sky. A large amount of galaxies images will be supplied by this space telescope and it will become critical to classify them with an automatic manner. As the images will be monochromatic, the classifier will have to deal with a unique channel. In this work, a new exploration way has been taken and it consists to use image processing in order to classify the galaxy types according to their morphology and shape. Only 2 galaxy types have been considered : elliptical and spiral galaxies. For this, a feature set has been built with wavelet and circular harmonics energies together with image moments and used to discriminate the galaxy types. Ultimately, our classifier (SVM : Support Vector Machine) obtained promising results and demonstrated the feasibility of this process. Furthermore, the experiment showed also that these features were robust against noise and additional objects on the images avoiding the preprocessing requirement (denoising, connected component labelling and so on). In future perspectives, one will try to add other feature kinds like DCT (Discrete Cosine Transform) or Fourier analysis for the improvement of the classification.