Gabor Convolutional Networks

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 endto-end pipeline.

Gabor Convolutional Networks

“…Finally, Section 5 concludes the paper with a few future research directions. The paper is the extension of our conference version [14].…”

GaborNet: Gabor filters with learnable parameters in deep convolutional neural network

“…A few works have explored Gabor filters for CNN. In [6] Gabor filters were used as preprocessing tool to generate Gabor features then using it as an input to a CNN, in [7] first or second layer of CNN was set as a constant Gabor filter bank, thus reducing number of trainable parameters of the network, and in [8] Convolutional Gabor orientation Filters were introduced, a special structure that modulates convolutional layers with learnable parameters by non-learnable Gabor filter bank. However, the authors did not report the integration of the filter parameters into the backpropagation algorithm.…”

Harmonic Networks with Limited Training Samples

“…Lu et al [20] designed a similar approach for medical image segmentation, however based on dual-tree complex wavelets. Robustness to scale and orientation of CNN is increased by modulating learned filters by a set of Gabor filters [21]. Rotation equivariance of learned features was accomplished by incorporated complex circular harmonics into CNNs [22].…”