An Improved LIC-based Pencil Drawing Simulation Method

Non-photorealistic rendering(NPR)technologies have made it a reality to render an arbitrary digital image into various artistic styles.Among them pencil drawing simulation has drawn tremendous attention from the computer graphics communities.How to simulate pencil drawings more realistically has become a research hotspot.Although with the aid of technical innovations pencil drawing simulation methods have made continuous progress,there still remain some shortcomings.First,existing methods can produce hatching graduation only in terms of stroke intensity,but cannot produce hatching graduation in stroke density.Second,the priori vector field lead to the directions of simulated hatching results not accord with the common rules in real pencil drawings.In order to address the above mentioned problems,we improve the conventional pencil drawing simulation method in this paper.The main contributions are listed as follows:(1)Filtering an image degraded by random binary white noise(RBWN)by line integral convolution(LIC)along a priori vector field can produce hatching graduation in terms of stroke intensity,but cannot produce hatching graduation in stroke density.We address this issue from a noise point of view by assessing several noise models and subsequently proposing a new noise model,called hybrid noise.The new noise model has been applied into LIC-based pencil hatching simulation method.In order to objectively assess hatching graduation results,we introduce a quantified graduality measurements,with which we demonstrate that the newly proposed hybrid noise outperforms the competitors.(2)In conventional LIC-based pencil hatching simulation method,the priori vector field is straight,leading to the directions of simulated hatching results not accord with the common rules in real pencil drawings.Therefore,after analyzing real hatching directions,we introduce a parabolic vector field to more realistically simulate hatching trajectory of shading.Combined with oblique vector field,we implement different hatching patterns and pencil grades.(3)The learning-based methods use deep neural networks for style transfer,but their pencil drawing results embrace disordered strokes and misty hatching.Hence we consider that combine the conventional LIC-based pencil hatching simulation method with the deep learning algorithm.We first propose the idea of learning hatching directions of real pencil drawings by neural networks.We segment the input images by trained neural networks and then use the segmentation to guide hatching directions.(4)To oversee the effectiveness of our improved LIC-based pencil drawing simulation method,we implement the whole pipeline of pencil drawing simulation.In the meantime,in order to enhance the artistic effect,we further extend our method into color pencil drawing simulation method and saliency-aware pencil drawing simulation method.In this paper,through theoretical analysis and a number of experiments we confirm the feasibility of our proposed method.Our method can produce hatching graduation not only in terms of stroke intensity but also in terms of stroke density.Moreover,we improve hatching directions,which helps us to generate more realistic simulated pencil drawings.Compared with existing methods,our method can produce visually more attractive results.