Peak detection strategy

Detecting enough peaks in any image in an automatic way is not easy to generalize. A strategy has been established to try to gather as many features as possible. Input parameters are: min_p and max_p, resp. minimum and maximum number of points to be detected, K the threshold factor as described above in the peak detector. In the following items, as soon as peak detection has returned a correct number of points we exit and return the found points. If not enough points have been found, we proceed to next step. If more points than requested have been found, we declare the search successful and return the max_p brightest points found in the image.

• Apply the peak detector with an initial value for K.
• Apply a low-pass filter to the image (e.g. a flat 5x5 kernel), apply the peak detector with the initial value for K.
• On the low-pass filtered image obtained at the previous step, try peak detection using K/2.
• On the low-pass filtered image, apply detection with K/5.
• If we reach this point, it means no list of points can be found in the input image. If max_p>1 return a failure code, else go ahead.
• If the image maximum pixel is usable (far enough from the edges to enable cross-correlation), return it.
• At this point, no point was found and max_p=1, issue a warning about the poor image quality and return the image center.

Normally, in a decent astronomical image there should always be sources to detect for this algorithm. The 3 parameters to set for object detection are specified in a specific section of the initialization file for the jitter program.

[DetPeak]
Threshold           = 1.0 ;
MinDetectPoints     = 1 ;
MaxDetectPoints     = 10 ;

Users can also provide their own lists of features through a configuration ASCII file, if they do not want to rely on this automatic feature extraction. Indeed, for images containing only faint sources (e.g. narrow-band imaging), the above strategy may fail to find anything but noise in the images, thus corrupt the offset measurements.