Prof. Dr. H. Dersch - HFU Furtwangen
Seam placement in panoramas is usually controlled by the panorama stitcher software. PTStitcherNG offers several modes: by default the center of the overlapping region between any two images is chosen. A second, smart mode is available which tries to place the seam depending on image content and which is not covered in this article. In any case, it is often desirable to manually move the seam position, e.g. to hide parts of the camera/tripod assembly, or to place the seam around moving objects.
Usually, this is done by either creating an appropriate alpha mask for the input image in question, or by editing the alpha channels of the intermediate, remapped images prior to feeding them into the blender. Besides these two,PTStitcherNG offers a third method which allows the user to create a reusable stitching mask. This mask encodes each source image as a unique color. Editing the mask is as simple as painting using any standard graphics program.
In the following we demonstrate this technique using a sample panorama project. It consists of four images taken with a fisheye lens and the horizontal and vertical positions indicated below. The original source image dimensions are 2592x3888 pixels.
|Image 1||Image 2||Image 3||Image 4|
|yaw = 0°||yaw = 180°||yaw = 90°||yaw = 270°|
|pitch = 20°||pitch = 20°||pitch = -20°||pitch = -20°|
As can be seen in images 3 and 4, parts of the panorama head and camera mounting bracket are visible. The area covered by the bracket (lower left portion of image 3 and 4) is visible in the opposite image. That means it should be possible to hide this part in the final panorama. Unfortunately, this is not the case in default stitching mode.
The images were aligned using PTGui and stitched using PTStitcherNG in default mode. As can be seen below, the seam runs straight through the bracket at both sides. And while the multiresolution blender does its best to blend the bracket with the background, it remains quite visible. The PTGui builtin stitcher yields virtually the same result requiring approximately 5 times as long.
Table of Contents
Copyright H. Dersch 2010