Abstract
We present an automated process for the morphological classification of galaxies in elliptical, spiral, and irregular types. The process can be divided in three stages: (1) Digital treatment of the image: which includes filtering (noise removal), segmentation (image centering and extraction), rotation (to a standard orientation), resizing (to a standard size), and decomposition (color, gray scale, or combined). (2) Parameter extraction: the galaxy is characterized by eigenvectors derived via principal component analysis. (3) Automatic learning using the following algorithms: artificial neural networks, K neighbor, regression locally weighed, and support vector machines, which produce the desired answer, the type of galaxy. The tests were developed with a group of 450 images with different galaxies types. A combination between the type becomes thef input image, the algorithm of extraction of parameters is fixed, and the algorithms of automatic learning are varied. The best combination of these algorthms provide an exactitude of 84%. Since the algorithm processes enormous amounts of data with thousands of calculations, the processing time is a problem. We are considering using parallel processing at various stages to improve this situation.
