A temporal memory-based ordered dithering using recurrent neural network

Abstract

Ordered dither is a popular digital halftoning technique used for bi-level displays. Despite its popularity, this technique results in two major problems e.g. objectionable periodic patterns and false contouring. This paper presents a temporal memory-based approach for ordered dithering. Digital halftoning has been considered as an optimisation problem. The temporal memory of recurrent neural network (RNN) has been employed to generate optimised halftone patterns. The temporal memory of RNN is used to retain association between pixels under processing and pixels already processed for generating the screen matrix. The human visual system (HVS) characteristics have been involved in cost function formulation. The proposed framework has been investigated with two most commonly practiced architectures of RNN, namely, Elman RNN and Jordan RNN. The results obtained using both the architectures have been presented pictorially. The results have also been evaluated against quality evaluation metrics such as peak signal-to-noise ratio (PSNR), universal quality index (UQI) and structural similarity index measure (SSIM). The results show that the proposed method may be potentially useful in the field of digital halftoning.