The novel index of the similarity between hand-drawn sketches for machine learning
The novel index of the similarity between hand-drawn sketches for machine learning
复制标题
DOI：
10.1109/isai-nlp48611.2019.9045433
复制
发表时间：
2019-10
期刊：
影响因子：
--
通讯作者：
Ryosuke Fujii;N. Mori;M. Okada
中科院分区：
文献类型：
--
作者：
Ryosuke Fujii;N. Mori;M. Okada
我们已经处理过手绘的草图。它们是包含人类情感和感性的创作。要在计算机上处理草图，有必要建立一个能对笔画顺序进行定量评估的指标。我们已经通过余弦相似度提出了一种针对笔触运动的草图相似度指标。然而，我们还发现，当人们理解所画内容时，人类的认知在很大程度上取决于草图的形状。出于这个原因，我们提出了一种新的相似度指标，它是在原有指标基础上增加了一个形状相似度指标。结果表明，由余弦相似度和结构相似性（SSIM）组成的相似度指标与人类对相似度的评估具有很强的正相关性，相关系数为0.7566。因此，我们定义了一个笔画顺序相似度的客观评估模型，该模型不仅包含草图的笔触运动，还包含其形状，并通过实验验证了其有效性。
We have dealt with sketches hand-drawn by human hands. They are creations including human emotions and sensibility. To handle sketches on a computer, it is necessary to have an index that quantitatively evaluates the stroke-order. We have already proposed a sketch similarity index for the pen movement by cosine similarity. However, we also found that human cognition strongly depends on the shape of a sketch when they understand what is drawn. For this reason, we propose a new similarity index, which is the old index added to a shape similarity index. As a result, the similarity index composed of cosine similarity and Structural Similarity (SSIM) has a strong positive correlation with the human evaluation of similarity. Then, the correlation coefficient is 0.7566. Therefore, we define an objective evaluation model for the stroke-order similarity that incorporates not only the pen movement of a sketch but also the shape of one, and we verify its effectiveness by our experiments.