Fusion of Complex Networks-Based Global and Local Features for Feature Representation

To realize effective feature representation, this paper proposes a new complex networks (CNs)-based feature fusion scheme to recognize texture images. Specifically, we propose two feature extractors to detect the global and local features of texture images respectively. To capture the global features, we first map a texture image as an undirected graph based on pixel location and intensity, and three feature measurements are selected to further decipher the image features, which retains as much image information as possible. Next, given the original band images (BIs) and the generated feature images, we encode them using local binary patterns (LBPs). Thus, the global feature vector is obtained by concatenating four spatial histograms. To decipher the local features, we jointly transfer and fine-tune a pre-trained VGGNet-16 model. After then, we fuse the middle outputs of max-pooling layers (MPs) and generate the local feature vector based on a global average pooling layer (GAP). Finally, the global and local feature vectors are connected to form the final feature representation of texture images. Experiment results show that the proposed scheme outperforms the state-of-the-art statistical descriptors and several deep convolutional neural network (CNN) models.