A High Quality Image Stitching Process forIndustrial Image Processing and QualityAssurance

The size of the recording area of a camera is limited.
The resolution of a camera image is also limited. To capture
larger areas, a wide angle lens can be used, for example. However,
the image resolution per unit area decreases. The decreased
image resolution can be compensated by image sensors with a
higher number of pixels. However, the use of a high pixel number
of image sensors is limited to the current state of the art and
availability of real image sensors. Furthermore the use of a wide
angle lens has the disadvantage of a stronger distortion of the
image scene. Also the viewing direction from a central location
is usually problematic in the outer areas of a wide angle lens.
Instead of using a wide angle lens, there is still the possibility
to capture the large image scene with several images. This can
be done either by moving the camera or by using several cameras
that are positioned accordingly. In case of multiple image
captures, the single use of the required image is a simple way
to evaluat e a limited area of a large image scene with image
processing. For example, it can be determined whether a feature
limited by the size is present in the image scene. The use
of this simple variant of a moving camera system or the use of
single images makes it difficult or even impossible to use some
image processing options. For example, determining the positions
and dimensions of features that exceed a single image is
difficult. With moving camera systems, the required mechanics
add to the effort, which is subject to wear and tear and introduces
a time factor. Image stitching techniques can reduce many
of these problems in large image scenes. Here, single images
are captured (by one or more cameras) and stitched together to
fit. The original smaller single images are merged into a larger
coherent image scene. Difficulties that arise here and are problematic
for the use in industrial image processing are, among
others: the exact positioning of the single images to each other
and the actual joining of the imag es, if possible without creating
disturbing artifacts. This publication is intended to make a
contribution to this.