Manifold Learning for Hand Drawn Sketches
Zhengyu Huang, Haoran Xie, Kazunori Miyata · 2020
It is a challenge issue to recognize and comprehend hand drawn sketches for various applications such as image retrieval and image-based modeling. In this work, we propose an unsupervised learning framework to obtain the manifold of hand drawn sketches. We use DCT (Discrete Cosine Transform) to exact the feature of preprocessed sketch images from Quick Draw Dataset and adopt LLP (Locality Preserving Projections) to calculate the 2-D manifold of these sketches. Experiment result in flower sketches demonstrates the proposed approach is suitable to represent the manifold of hand drawn sketch.