Deep Neural Inverse Halftoning

Inverse halftoning is a kind of technology which transforms binary images composed of black and white pixels to continuous-tone images. Many scholars have studied this problem so far, but the results are not satisfactory. In this paper, we propose an end-to-end deep convolutional neural network composed of two parts. The first part is the feature extraction part which consists of 4 convolution layers and 4 pooling layers to extract feature from the halftoning images. The second part is the reconstruction part which contains 4 deconvolution layers to reconstruct the continuous-tone images. A U-Net structure which concatenates the outputs from the feature extraction layers with deconvolution layers is used for better restoring the detail information of the original images. Experimental results show that our method outperforms the state-of-arts in terms of both visual quality and numerical evaluation