In the area of large-scale graph data representation and semi-supervised learning, deep graph-based convolutional neural networks have been widely applied. However, typical graph convolutional network (GCN) aggregates information of neighbor nodes based on binary neighborhood similarity (adjacency matrix). It treats all neighbor nodes of one node equally, which does not suppress the influence of dissimilar neighbor nodes. In this paper, we investigate GCN based on similarity matrix instead of adjacency matrix of graph nodes. Gaussian heat kernel similarity in Euclidean space is first adopted, which is named EGCN. Then biologically inspired manifold similarity is trained in reproducing kernel Hilbert space (RKHS), based on which a manifold GCN (named MGCN) is proposed for graph data representation and semi-supervised learning with four different kernel types. The proposed method is evaluated with extensive experiments on four benchmark document citation network datasets. The objective function of manifold similarity learning converges very quickly on different datasets using various kernel functions. Compared with state-of-the-art methods, our method is very competitive in terms of graph node recognition accuracy. In particular, the recognition rates of MGCN (Gaussian kernel) and MGCN (Polynomial Kernel) outperform that of typical GCN about 3.8% on Cora dataset, 3.5% on Citeseer dataset, 1.3% on Pubmed dataset and 4% on Cora_ML dataset, respectively. Although the proposed MGCN is relatively simple and easy to implement, it can discover local manifold structure by manifold similarity learning and suppress the influence of dissimilar neighbor nodes, which shows the effectiveness of the proposed MGCN.
In recent years, iris recognition has been widely used in various fields. As the first step of iris recognition, segmentation accuracy is of great significance to the final recognition. However, iris images exhibit a variety of noise in the real world, which leads to lower segmentation accuracy than the ideal case. To address this problem, this paper proposes an iris segmentation method using feature channel optimization for noisy images. The method for non-ideal environments with noise is more suitable for practical applications. We add dense blocks and dilated convolutional layers to the encoder so that the information gradient flow obtained by different layers can be reused, and the receptive field can be expanded. In the decoder, based on Jensen-Shannon (JS) divergence, we first recalculate the weight of the feature channels obtained from each layer, which enhances the useful information and suppresses the interference information in the noisy environments to boost the segmentation accuracy. The proposed architecture is validated in the CASIA v4.0 interval (CASIA) and IIT Delhi v1.0 datasets (IITD). For CASIA, the mean error rate is 0.78%, and the F-measure value is 98.21%. For IITD, the mean error rate is 0.97%, and the F-measure value is 97.87%. Experimental results show that the proposed method outperforms other state-of-art methods under noisy environments, such as Gaussian blur, Gaussian noise, and salt and pepper noise.
