Deep Gabor Scattering Network for Image Classification.

Deep learning models obtain exponential ascension in the field of image classification in recent years, and have become the most active research branch in AI research. The success of deep learning prompts us to make greater achievements in image classification. How to obtain effective feature representation becomes particularly important. In this paper, we combine the wavelet transformation and the idea of deep learning to build a new deep learning model, called Deep Gabor Scattering Network (DGSN). Concretely, in DGSN, we use the Gabor wavelet transformation to extract the invariant information of the images, partial least square regression (PLSR) for feature selection, and support vector machine (SVM) for classification. A key benefit of DGSN is that Gabor wavelet transformation can extract rich invariant features from the images. We show that DGSN is computationally simpler and delivers higher classification accuracy than related methods.