Search

Image synthesis and super-resolution (SR) have always been a hot spot for computer vision and image processing research. Since the development of Deep Learning, especially after the Deep Convolutional Generative Adversarial Network (DC-GAN) methods, facial image synthesis and SR problem had been solved in many circumstances. But most of the existing works were focused on natural-looking of the synthesized result rather than keeping facial information of the original image. Our paper presented an end-to-end method of getting high-resolution photo-realistic facial images from low-resolution (LR) in-the-wild images without losing the facial identity details. The pipeline used a flexible stacked GAN structure for the SR process with different target image resolutions on different upscaling factors. To avoid getting blur or nonsensical image output and realize the flexibility, “U-Net” architecture and upsampling layers with residual learning blocks were stacked. The stacked network structure makes applying different loss functions in different parts of the network possible, which helps to solve the two problems of keeping identical facial details of the LR input image and generating high-quality output images simultaneously. By using 3 different loss functions in different positions of the stacked network separately, through experimental comparison, we found the best stacked residual block parameters which could get the best output image quality. Experimental results also explicated that the network had a good SR ability compare to state of the art methods in different resolution and upscaling factor.

Liquid computing is an effective approach to intelligent computations of neural networks, especially for spiking neural networks. If the liquid network is embedded with a proper structure it can perform complex computational tasks. However, the modeling of self-organized neural networks with more biological characteristics is still an important open challenge, resulting in major constraints on improving the computational capability and dynamical diversity of the model. Here, we present a novel type of liquid computing model with both multi-clustered and active-neuron-dominant structure of spiking neural network, instead of the traditional random structure. The optimal parameter settings of the cluster number and time window size of clustering generation method had been considered. The synaptic weights in each cluster are further refined through the spike-timing-dependent plasticity rule to obtain an active-neuron-dominant structure. The results show that this model has much better performance on liquid computing than the random model. The enhancement of information processing capability is achieved by improving two aspects, i.e. the activity synchrony and network sensitivity, based on the clustered structure and active-neuron-dominant structure, respectively. Statistical analysis demonstrates that both structure entropy and activity entropy of our proposed network are increased, indicating its high topological complexity and dynamical diversity. Therefore, the improvement in efficiency of signal transmission of this network is confirmed.