A Segment-Based Disparity Estimation Algorithm for Multiview Image Interpolation

• Main Authors: Yu-ChenHung, 洪譽禎
• Other Authors: Jar-Ferr Yang
• Format: Others
• Language: en_US
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/91037534857093146325

碩士 === 國立成功大學 === 電腦與通信工程研究所 === 98 === Researches in both computer vision and graphics have made great progress toward virtual image generation. Virtual image generation is applied to FTV which is an innovative application that permits the viewers selecting the desired viewpoints. Supposing the real images captured by multiple cameras at predefined viewpoints are available, the virtual images are able to be generated by interpolating the captured real images. The objective of the thesis is to realize a virtual image generating system. In the thesis, a segment-based disparity estimation algorithm is proposed first. Block-based segmentation which is not tedious and time-consuming is utilized. The scheme for disparity refinement is also presented. Next, three image interpolation approaches are described in details. 3D stereo images are able to be synthesized using the proposed interpolation method. Finally, hole filling is required for the purpose of obtaining the delicate interpolated images.