
Tai-Hsu Lin
Department of Electrical Engineering (formerly)
National Taiwan University
e-mail: lord.lth@gmail.com
Facebook: Tai-Hsu Shecky Lin

Department of Electrical Engineering (formerly)
National Taiwan University
e-mail: lord.lth@gmail.com
Facebook: Tai-Hsu Shecky Lin
| HDR Image Acquisition VFX@ NTU, 2008 Project 1 |
|
| We implemented a program which automatically recovers a high dynamic range image from photos captured with different exposure times. Keywords: MTB registration, Paul Devebec's method, Fast bilateral filter, White balance. |
| Panorama Stitching VFX@ NTU, 2008 Project 2 |
|
| We implemented a program which automatically extracts and stitches panoramas from a photo collection. Keywords: SIFT, feature matching, RANSAC, Homography, Bundle adjustment, Multi-band blending. |
| "Texture-Aware" Image Resizing VFX@ NTU, 2008 Final Project |
|
| We implemented a content-aware image resizing algorithm based on the concept of "Bidirectional Similarity." We have also tried to improve the algorithm with the idea from the texture synthesis field. Keywords: Bidirectional similarity, Content-aware image resizing, Appearance transform, Dimension reduction, Nearest-neighbor search, Texture synthesis. |
| Paper Review Blog AMMAI@ NTU, 2008 |
|
| This is a blog where I posted review comments for the reading materials from the class. Most of them are on machine learning, image/video retrieval and pattern recognition. |
| Face Morphing Digital Image Processing@ NTU, 2007 Final Project |
|
| We implemented a software which performs the face morphing effect. The program can run automatically or according to user controls. Keywords: Face detection, Active appearance model, Mesh warping, Laplacian, MRF stitching. |
Work Projects & materials
| Automatic Optical Inspection System Test Research Inc., Fall 2007 - Summer 2008 |
|
| We developed a high-speed automated optical inspection system for PCB boards. The program transforms the target images into feature points and searches it in the input subject to translations and rotations. It could run in a few hundrea milliseconds on mega-pixel images with a precision higher than 98% and provide sub-pixel accuracy. |