Sketch-based 3D modeling of symmetric objects from wireframe sketches on paper

• Main Author: Yahya, Zahrah
• Format: Thesis
• Language: English
• Published: 2016
• Subjects: Computer drawing; User interfaces (Computer systems); Computer-aided design
• Online Access: http://psasir.upm.edu.my/id/eprint/69321/1/FSKTM%202016%209%20IR.pdf; http://psasir.upm.edu.my/id/eprint/69321/
• Institution: Universiti Putra Malaysia

The motivation of this work originates from recent studies on sketch-based modeling that emphasize on the reliance of the designers on traditional methods for conveying their ideas. Many studies have chosen to focus on 3D modeling from digital sketching because of the challenges and complexity of reconstructing from paper sketches. Considering the complications of paper sketching, it is still an ideation strategy proven as an effective, flexible and instantaneous tool compared to sketching with computer-based instruments. Later, this ideation is needed to be visualized in three dimensions for better understanding. Therefore, an automated application is required to transform paper sketches into three-dimensional models. This application should constitute the essential processes for a computer to understand the paper sketch junctions, depth and connections. Feature extraction is a preliminary step before 3D reconstruction and therefore, a junction detection algorithm is proposed. The presented algorithm benefits in omitting unnecessary junctions that are problematic for object recognition in comparison to the closest algorithm to our work. The algorithm is validated by conducting manual ground truth and the results were compared with the closest and well-known methods. A high accuracy of 99.42% and a correctness of 97% have been achieved. The proposed method shows an optimum result while resolving the existing limitations. Subsequently, object face dentification is a common step to provide important information for 3D reconstruction. The proposed face identification method is based on symmetry regularities in determining the faces of the object. The previous works employed both edges and vertex properties while using a new perspective and based on our studies we understood that we can solely rely on vertex representation for the identification process. By neglecting the edges, less ambiguity is expected but the work will be open to new challenges due to the need to preserve high accuracy. The results of the validation show a complete similarity between each original face and the output from our algorithm. Furthermore, the algorithm is also evaluated by human experts and shows 100% correctness.