help on image processing, image segmentation, pattern recognition and contour tracking

Dear all,

I have to process a stack of 5000 images and locate droplets as well as cloud structures in these images. I did a lot of image processing with Matlab but all the techniques I applied are not very reliable to extract the features I am interested in out of the images. I put two sample images of my measurements on a website so that you get an idea of what I am talking about. Feel free to download the images if you like. I also created some movies of the images that can be found on the same website.

Link to website:

http://wwwuser.gwdg.de/~fwinkel/images.html

1. for the images with only the droplets:

In the images with only the droplets I want to find the center of each droplet and track it in time to determine the fall-down period. The problem is how to exactly identify and find the center of each droplet. The intensity of a droplet is varying over time because they continuously develop, grow and eventually fall down from the top plate. The artefact that a droplet consists of a dark and a bright region is caused by the method of illumination. The images are already background corrected. In the beginning when a droplet starts to grow the intensity is only slightly different compared to the background intensity of the image. With time the droplet gets bigger, scatters more light and becomes darker until it eventually falls down. The advantage is that a droplet always develops at the same position and all the droplets have the same diameter. To analyse the images I already calculated the gradient in x-direction (looks exactly like the 2nd derivative of a gaussian) then did a 2d-convolution with a sample function to point out the position of each droplet. Then I used a peak-finding algorithm to locate the center of each droplet in the convolved image and a particle tracking algorithm to track the droplets in time. However the peak-finding algorithm is not very reliable. Sometimes wrong peak locations are detected or droplets are not recognized. So my method does not work very well at the moment. For the analysis I only considered a 400pxl x 400pxl region of the image. Maybe someone has a good idea/suggestion or knows a method of how to exactly locate each droplet in a certain region of the image.

2. for the images with the clouds:

In the images with the clouds I want to separate the clouds from the rest of the image to determine the area and the perimeter of the cloud regions and track these features in time. The clouds also show a honeycomb or cell structure (thin dark lines). Maybe it is possible to get a skeletonized view of these structures as well. The problem is that the droplets are also visbile in the cloud images. I think it is not possible to filter the droplets out of the image. I already lowpass-filtered the images (2d-fft) and then applied a 2d active contour tracking algorithm (ChanVese) to extract roughly the shape of the clouds but this method is not very reliable. Maybe someone has a better idea of how to overcome this problem or how to remove the droplets from the image so that the cloud structures can be extracted and analysed.

I greatly appreciate any help. Many thanks in advance.

Best wishes, Florian