When conducting industrial #research, images of the evolving process can provide a wealth of information about the system under examination.
One technique to identify key features within the images, is to use convolutional operations. An advantage of convolutional filters is that they are spatially invariant. Imagine that you are trying to find a specific type of feature in an image such as an edge, ridge, or more complex structure.
To find this feature in a given patch of the image you can use a filter tailored to the type of feature you want to detect. The filter itself is usually a mask that is the same size as the patch to which it is applied, corresponding to the characteristic size of the feature of interest.
When the feature of interest is present in the patch the filter gives a strong response. The convolutional operation applies the filter to the whole image by sliding it across the image and measuring the response at each position. The result is a new image of the filter responses, corresponding to the feature of interest.
Our automated droplet detection #algorithm uses a Hessian ridge filter, to help locate oil-water (or air-water) interfaces and the outlines of droplets.
#imageprocessing #imageanalysis #imagerecognition