Gabor Convolutional Networks

TL;DR

Gabor Convolutional Networks integrate Gabor filters into deep CNNs to improve robustness against orientation and scale variations, achieving high recognition accuracy with fewer parameters.

Contribution

This paper introduces Gabor CNNs, a novel deep learning model that incorporates Gabor filters into CNNs to enhance invariance to transformations while reducing parameter count.

Findings

Superior object recognition under scale and rotation changes

Fewer learnable parameters compared to traditional CNNs

Easy integration with existing deep learning architectures

Abstract

Steerable properties dominate the design of traditional filters, e.g., Gabor filters, and endow features the capability of dealing with spatial transformations. However, such excellent properties have not been well explored in the popular deep convolutional neural networks (DCNNs). In this paper, we propose a new deep model, termed Gabor Convolutional Networks (GCNs or Gabor CNNs), which incorporates Gabor filters into DCNNs to enhance the resistance of deep learned features to the orientation and scale changes. By only manipulating the basic element of DCNNs based on Gabor filters, i.e., the convolution operator, GCNs can be easily implemented and are compatible with any popular deep learning architecture. Experimental results demonstrate the super capability of our algorithm in recognizing objects, where the scale and rotation changes occur frequently. The proposed GCNs have much fewer learnable network parameters, and thus is easier to train with an end-to-end pipeline.