| Summary: | 碩士 === 國立中正大學 === 資訊工程所 === 98 === In this thesis, a low-complexity image stitching is proposed. User can automatically merge images with overlapped area into a panoramic image, and even these source images have rotation and zooming actions.
In image stitching processing, how to get a correct image alignment is the most important and necessary no matter using the camera parameters or utilizing the original resource like color or texture. In existing methods, using camera parameters is the easiest way to implement. However, it is not suitable for most people since it is not so easy to get the camera parameters. Compared with camera parameters, utilizing a linear transformation to merge images into a panoramic image is more suitable.
In our design, we use feature-base estimation to get the corresponding points between images and employ down-sampling method in finding the feature points to reduce computational complexity. We also apply a copy-paste and texture-base scheme to remove the defects in resulted image.
The performance is effective enough to implement on an embedded platform. Compared with common personal computer software tools, Photoshop, our design is about 2 times faster, and it is also about 3 times fast than the other designs. Because of the low-complexity and flow in our design, it can be extended to stitch a wide-of-video.
|
|---|
