| Summary: | 碩士 === 國立中興大學 === 資訊科學與工程學系 === 105 === The panorama images creating from stitching technique mostly have irregular boundaries. Photographer and common users generally prefer that panorama images have rectangular boundaries. To get this result, we can use the rectangling method like cropping or inpainting, but cropping method will lose the information outside the chosen inner rectangle and inpainting method will use the content that choose from the current image to fill up the missing area. These method will reduce or destroy the wide view of original panorama images. With this problem, the method of using warping to rectangling panorama images is developed.
In this paper, we present a new rectangling method for panorama images using scale-and-stretch-based content-aware warping. The original warping method for panorama images consists of two steps. The first local step use mesh-free local warping to generate the irregular mesh on the input image. The second global step optimizes this mesh to get the rectangle result. Our proposed method modifies the global step based on significance map to allow diverting the distortion due to rectangling to unimportant area, such that the impact on perceptually important features is minimized. In order to protect more content, we modify the original significance map to get more effective information. In the experimental results, our proposed method can get the better result with the scale-and-stretch-based content-aware warping.
|
|---|
