Two-channel Convolutional Neural Network Research Articles

Medical Hyperspectral Image Classification Based on End-to-End Fusion Deep Neural Network

To solve the problem of supervised convolutional neural network (CNN) models suffering from limited samples, a two-channel CNN is developed for medical hyperspectral images (MHSI) classification tasks. In the proposed network, one channel of end-to-end network, denoted as EtoE-Net, is designed to realize unsupervised learning, obtaining representative and global fused features with fewer noises, by building pixel-by-pixel mapping between the two source data, i.e., the original MHSI data and its principal component. On the other hand, a simple but efficient CNN is employed to supply local detailed information. The features extracted from different underlying layers of two channels (i.e., EtoE-Net and typical CNN) are concatenated into a vector, which is expected to preserve global and local informations simultaneously. Furthermore, the two-channel deep fusion network, named as EtoE-Fusion, is designed, where the full connection is employed for feature dimensionality reduction. To evaluate the effectiveness of the proposed framework, experiments on two MHSI data sets are implemented, and results confirm the potentiality of the proposed method in MHSI classification.

Two-channel convolutional neural network for facial expression recognition using facial parts

This paper proposes the design of a facial expression recognition system based on the deep convolutional neural network by using facial parts. In this work, a solution for facial expression recognition that uses a combination of algorithms for face detection, feature extraction and classification is discussed. The proposed method builds a two-channel convolutional neural network model in which facial parts are used as inputs: the extracted eyes are used as inputs to the first channel, while the mouth is the input into the second channel. Feature information from both channels converges in a fully connected layer which is used to learn global information from these local features and is then used for expression classification. Experiments are carried out on the Japanese female facial expression dataset and the extended Cohn-Kanada dataset to determine the expression recognition accuracy for the proposed facial expression recognition system based on facial parts. The results achieved shows that the system provides state of art classification accuracy with 97.6% and 95.7% respectively when compared to other methods.

Stereoscopic image quality assessment by deep convolutional neural network

In this paper, we propose a no-reference (NR) quality assessment method for stereoscopic images by deep convolutional neural network (DCNN). Inspired by the internal generative mechanism (IGM) in the human brain, which shows that the brain first analyzes the perceptual information and then extract effective visual information. Meanwhile, in order to simulate the inner interaction process in the human visual system (HVS) when perceiving the visual quality of stereoscopic images, we construct a two-channel DCNN to evaluate the visual quality of stereoscopic images. First, we design a Siamese Network to extract high-level semantic features of left- and right-view images for simulating the process of information extraction in the brain. Second, to imitate the information interaction process in the HVS, we combine the high-level features of left- and right-view images by convolutional operations. Finally, the information after interactive processing is used to estimate the visual quality of stereoscopic image. Experimental results show that the proposed method can estimate the visual quality of stereoscopic images accurately, which also demonstrate the effectiveness of the proposed two-channel convolutional neural network in simulating the perception mechanism in the HVS.

基于双通道卷积神经网络的深度图超分辨研究

基于双通道卷积神经网络的深度图超分辨研究

A two-channel convolutional neural network for image super-resolution

Abstract A two-channel convolutional neural network (including one shallow and one deep channel) is proposed for the single image super-resolution (SISR). Most existing methods based on convolution neural networks (CNNs) for super resolution have a shallow channel which easily loses the detailed information. And these methods need preprocessing such as bicubic interpolation enlarging LR images to the size of HR images, which may introduce new noises. Meanwhile, most of them use only one fixed filter during the reconstruction. The proposed algorithm solves the above problems, which is named shallow and deep convolutional networks for image super-resolution (SDSR). First, the proposed method uses two channels: shallow and deep channel. The shallow channel mainly restores the general outline of the image. On the contrast, the deep channel extracts the detailed texture information. Second, the proposed method directly learns an end-to-end mapping between low-resolution (LR) and high-resolution (HR) images, which does not need hand-designed preprocessing. The upsampling of the network by deconvolution is embedded in the two channels, which leads to much more efficient and effective training, reducing the computational complexity of the overall SR operation. Finally, during the last period of reconstruction, the deep channel adopts multi-scale manner, which can extract both the short- and long-scale texture information simultaneously. Our model is evaluated on the different public datasets including images and videos. Experimental results demonstrate that the proposed method outperforms the existing methods in accuracy and visual impression.