The correlation between the raw image and the chip or georeferenced image is performed in three progressive levels, each using a predefined kernel that defines the size of the area used to calculate the correlation score.
Matching begins on a low-resolution version of the image. The kernel starts at a location based on the project's approximate georeferencing and moves over a defined search area to establish an approximate match. The process repeats using a progressively smaller kernel on a medium-resolution version of the image and then on a full resolution version to produce the most precise match possible. The correlation score for the full resolution match appears in the Automatic GCP Collection window under Collected GCPs in the Score column. A high correlation score usually means that the identified matching features are a successful match. Scores lower than the minimum acceptance threshold are shown in red and are marked as failed.
When you decide whether to use a GCP or not, do not base your decision solely on the pass or fail status. The correlation scores are also a valuable tool to use to evaluate the quality of the GCP. A perfect match may still have a low correlation score since the chip and the raw image may have been taken at slightly different resolutions, in different illumination conditions, and with different sensors. On the contrary, repetitive features, such as lines in a parking lot, can produce a perfect correlation score because the features look identical, but the match may be on the wrong feature.
The search radius defines the space around the kernel in which it will search for matching features. For example, if the search radius is 100 with search units set to pixels, the search area will cover 100 pixels above, below, left, and right of the kernel.
Because the location for the first level search is a best guess based on the project's approximate georeferencing, the kernel might not cover the best matching feature. By creating a search zone in the immediate vicinity of the kernel, you increase the probability of finding a perfect match.
In most cases, the default search radius and units of 100 pixels is sufficient since the process will generally start with a fairly good approximate location. For projects using high resolution images or containing poorly estimated math model solutions, the first guess at the location may be many pixels away from the matching feature. To improve the results you can use a larger search radius, which will increase the computation time, or you can improve the first guess by improving the initial math model by adding GCPs or obtaining better exterior orientation data, for example. If you have a very good initial math model and 'first guess' location, you can use a smaller search radius, which will decrease computation time and reduce the chances of a mismatch.
© PCI Geomatics Enterprises, Inc.®, 2024. All rights reserved.