Quantitative Evaluation of Line Thickness in AI-Generated Anime Line Art Using Image Processing

Recent advances in diffusion-based generative models and conditional generation techniques have enabled the widespread use of automatic line art generation from anime-style images. However, AI-generated line art often exhibits spatially inconsistent line thickness, which poses challenges for objective quality evaluation and systematic control. Although variations in line thickness directly affect visual quality, there remains a lack of well-established methods for quantitatively analyzing this property. In this paper, we propose an image processing–based method for the quantitative evaluation of line thickness in AI-generated anime-style line art. By leveraging distance transform and the Local Thickness algorithm, line thickness is measured as pixel-level distance values, and the distribution characteristics are analyzed using the mean and standard deviation as key metrics. The proposed method is independent of the underlying line art generation model and enables numerical representation of local thickness variations within line drawings. We conduct both single-image distribution analysis and multi-image statistical analysis on a dataset of 100 line art images. The single-image analysis demonstrates that line thickness should be interpreted as a distribution rather than a single scalar value. Furthermore, scatter plot analysis across multiple images reveals diverse patterns of average thickness and variability, indicating that mean values alone are insufficient to fully characterize line thickness properties. This work provides an objective, quantitative evaluation framework for analyzing line thickness in AI-generated line art. The proposed method can serve as a foundation for line art quality assessment, understanding the characteristics of training data, and establishing practical quality control indicators in animation and content production pipelines.