The discussion about NPP / LPP presented on this page is based on a simplified principle. For instance, on the 4-shots panorama given as an example, the seams (the stitching lines) between two adjacent and overlapping images are supposed to be equidistant from the center line of the individual images. In fact this is not the case because the stitching program itself complemented by a blending program (PTGui own blender engine, Enblend, SmartBlend, Photoshop auto-blend layers, etc.) make and place the seams at an optimized location that is in general far from the simple equidistant scheme.
We have recommended the LPP to be selected appropriately such as to minimize the errors along the (equidistant) seam. Then, at the last stage of the process, the blender places the actual and final seam for away from this position!
Let's show that with an illustration based on a 4-shot panorama...
The pixels on the "expected" seam lines should be from 45° minimum (on the horizontal) up to 90° maximum at the Nadir and Zenith.
Consequently this is the equirectangular output image that I thought we should have had:
These are the expected shape and location of the seams (the equidistant seams are shown as vertical red lines).
These are the seams border lines that are computed by PTGui Pro 9.1.2 and viewed in the preview windows: in some places the red line is quite different from expected.
|These are the first pair of stitched images (#1 and #3 after blending)||These are the second pair of stitched images (#2 and #4 after blending)|
Auto-blending layers by Photoshop CS4 was used for the demonstration (unlike others, it can output individual "blended" image layers with masks). The actual seams location and shape are obviously quite different from the expected seams (in red on the precedent image above). The angular distance from the center of the individual images are shown and labeled in degrees.
This is the panorama (Flash; 2.2 MB) with the vertical lines showing the "expected" seams in black.
The stitching lines and the seams after blending is complete are completely different and very far from the expectation. And this is absolutely normal. Blending is aiming at visually suppressing the seams altogether. The very principle of this function makes the limits that are actually built between the two images to be tortuous and over-complicated. But one has to know that the blending may also being working hard (and most often successfully) to suppress or lessen the errors due to parallax!
Most of the blenders can be finely tuned with command settings that shall strongly influence the shape of the limit between two stitched images. Another observation can be made: the order of introduction of the input source images has an strong influence on the shape of the same limit lines.
So should we then just forget about LPP, NPP careful adjustment? NO, NO, NO: the less paraxial induced stitching errors there are on the source images, the better shall be the cleaning of the seams that will be done by the blender. That's simple...