Scientific publications, especially biomedical publications, contain a large number of compound figures, which are composed of multiple graphs, plots, and drawings. With the growing interest in data mining, scientific image understanding, and retrieval, compound figure separation and label recognition have become vital steps for various downstream tasks. However, existing studies are difficult to apply to increasingly complex scenarios, and they usually treat these two tasks separately. In this work, we propose a new model called YOLO-OCR to do compound figure separation and label recognition simultaneously. The YOLO-OCR realizes object detection, text detection, and text recognition altogether in a unified end-to-end trainable network. Benefiting from shared convolution features, the model has fewer computation costs and higher performance. To reduce annotation costing, we train the model on a synthesized compound figure dataset and then finetune the model in actual compound figure datasets based on an active learning strategy. The results show that the proposed method achieves a new state-of-the-art performance on the ImageCLEF 2016 dataset and our dataset. In addition, we developed an online system based on the proposed model to help researchers conveniently separate compound figures. The project is publicly available at https://www.chatfigures.com/figure-separation.

The appearance of woven fabrics is intrinsically determined by the geometric details of their meso/micro scale structure. In this paper, we propose a multiscale representation and tessellation approach for woven fabrics. We extend the Displaced Subdivision Surface (DSS) to a representation named Interlaced/Intertwisted Displacement Subdivision Surface (IDSS). IDSS maps the geometric detail, scale by scale, onto a ternary interpolatory subdivision surface that is approximated by Bezier patches. This approach is designed for woven fabric rendering on DX11 GPUs. We introduce the Woven Patch, a structure based on DirectX’s new primitive, patch, to describe an area of a woven fabric so that it can be easily implemented in the graphics pipeline using a hull shader, a tessellator and a domain shader. We can render a woven piece of fabric at 25 frames per second on a low-performance NVIDIA 8400 MG mobile GPU. This allows for large-scale representations of woven fabrics that maintain the geometric variances of real yarn and fiber.

We propose a positional constraint method to solve the multi-layered deformable surface problem based on a master-slave scheme. This allows two or more deformable surfaces to be attached together in any orientation relative to each other for the purpose of modeling cloth attachments and multi-layered clothing. The method does not require the mesh resolution of the deformable surfaces to be the same or the matching of anchor points between layers. After the attachment process, the surfaces are treated as a multi-layered surface. However, this surface contains non-manifold features. We introduce a technique for preventing self-intersection of the non-manifold features. We demonstrate the stability of this method by performing several experiments with high surface complexity and a large number of colliding feature pairs. Interactive rates can easily be achieved for multilayered surfaces with an appropriate discretization level of triangles.