Digital Multitone Image Reconstruction using Deep Generative Adversarial Nets

Digital multitoning is a latest version of the halftone technology which is widely used in printing applications. The method transforms the original gray scale image into its appropriate printable version using more tones than conventional halftones. To be specific, the gray scale or color image is converted to its 2 or 3-bit version. The multitone image results in enhanced homogeneity, smooth texture and less halftone artifacts. On the other hand, many applications require reconstruction of printed images for digital processing and in most cases source image may not be available. In this paper, a deep learning model is proposed based on the conditional general adversarial nets (cGAN's) to perform effective reconstruction of multitone images. For experimentation, a digital multitone dataset is developed comprising of 10,000 images. The proposed model is optimized to perform with minimal layers and reduced parameters. The discriminator is constructed as Patch -GAN architechture and is implemented over the patch size of 64×64. From the overall experimental results, it has been validated that the proposed model achieves consistent and best reconstruction quality.