Saturday, 20 September 2008

Panorama Stitching

Since my last report on stitching panoramas, I have been out and taken a number of new sequences. I am discovering a range of issues that influence the quality of the final panorama. Some of these relate to the technique used in taking the images that compose the final image. Other relate to the stitching process itself.

When I go out to take an image, I now try and set the camera to manual for all settings. The reason is quite simple. Any change in focus or exposure can cause problems in the final stitched image. The same applies for adjustments in the white balance. I had mainly thought that this applied when taking photos under artificial lighting when the light temperature is lower and you are left with an orange tinge in the photos. However when taking a sequence of photos that make up the spherical image, the camera if set on auto white balance makes decisions about the nature of the light temperature and adjusts the image colours as it believes are required. The solution is simple. Never leave anything to be set automatically by the camera. Take control and set everything yourself.

Some of the stitching programs can compensate for some of these problems but not all. If the image is out of focus then it is out of focus. No attempt to improve it using software will resolve the problem. A change in focal length can be handled by a stitching program such as Hugin but if the change is too great then you might have gaps between the images. The stitching program can't full in the gaps. The panotools that are used by Hugin will also attempt to correct exposure problems by attempting to match the histogram of the the portions that overlap and adjusting one of the images to that revised exposure. However, this can lead to other problems. In effect, the less patch up work that has to be done in the processing work flow, the better.

Adjusting the camera settings to manual doesn't solve the change in light conditions that occurs when taking an image while the sun is setting (see the Days Bay wharf panorama). As the light fades, there is a need to adjust for the change in tones caused by the change in the amount of light available and possibly the change in colour of the light (i.e. a change in the light temperature). This is difficult to achieve manually and the automatic options simply compound the problem. With the Days Bay wharf sunset, the problem can be seen in the nadir image at the top of the sphere. It simply doesn't seem to match. I need to explore the blending options because I am not sure that these have been applied in a way that might resolve the problem.

The second issue in taking these images are movements in the camera positioning. I ensure that the tripod is level and endeavour to ensure that it won't move but even leaving your hand resting on the camera can change the camera angle by one or two degrees. This seems to have more impact when this angle change in the vertical dimension. Hugin provides a facility to find control points but the default option failed to find points between layers. There are also alternative programs that can be used from within Hugin and it can be time consuming testing each of these options. You also land up with a clutter of files that contain the definitions of the control points between the images.

As someone who wants to focus on the photography and not the processing after the event, I find this quite frustrating.

No comments: