[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Object recognition

Bertram Tenner wrote:
I simply have an image with unregularily distributed
Gaussian spots and, giving some input parameters (threshold, size etc), need
their number and approximate position.


If you have access to the new version of IDL (5.3), you could try the MORPH_TOPHAT function.   I haven't tried to implement
the tophat function in IDL5.2 yet but I want to.

My understanding of the tophat function is that it is used to extract features that are smaller and brighter than parameters you specify.  You might have to do some morphological manipulation (open and closing, smoothing,...) on the image first to clean up the background.

On the other hand, if you have a systematic and objective  method for determining a global threshold for each image, you could threshold the images (force the background to zero)and then locate your objects with the LABEL_REGION function.    Much depends on how well the image has been flat-fielded so that the background is uniform.  Global thresholding can be problematic if the background is is not uniform (brighter on one side than the other) or if the objects tend to overlap/merge or the objects you are looking for vary in brigtness.


Ben Tupper

Bigelow Laboratory for Ocean Science