Comparing Neural Style Transfer and Gradient-Based Algorithms in Brushstroke Rendering Tasks

Non-photorealistic rendering (NPR) with explicit brushstroke representation is essential for both high-grade imitating of artistic paintings and generating commands artistically skilled robots. Some algorithms this purpose have been recently developed based on simple heuristics, e.g., using an image gradient driving orientation. The notable drawback such the impossibility automatic learning to reproduce individual artist’s style. In contrast, popular neural style transfer (NST) are aimed at goal by their design. question arises: how good performance methods in comparison heuristic approaches? To answer question, we develop a novel method experimentally quantifying algorithms. This correlation analysis applied histograms six parameters: length, orientation, straightness, number neighboring brushstrokes (NBS-NB), similar orientations neighborhood (NBS-SO), orientation standard deviation (OSD-NB). numerically captures similarities differences distributions parameters allows two NPR We perform investigation generated algorithm NST algorithm. results imply that while give rather different parameter histograms, capabilities mimicking manner limited comparably. A direct NBS-NB these extracted from real painting confirms finding.