Seeking Assistance: Image analysis strategies


I have a problem to solve: We are placing stud bumps on a set of traces and I am looking to quantify the placement accuracy. I know the overall position and have good info on the position of the traces so I am working on the "bumps". Each part has >50,000 bumps and we need some thing that can at least do a reasonable sampling so any manual component is really unacceptable.

The question I need to answer is what is the XY center of the bump size and shape are not important. They also tend to be very similar in size which helps.

The image quality is limited and they have variability and there are some idiosyncrasies in their individual appearance including an inner ring or partial ring. The other problem the placement could be so bad as to be hanging off the trace and these are the most important ones to find. I have tried image particle analysis with various conditions including erosion and dilation, but the algorithms are not very robust. Additionally the lighting tends to skew the center of the "particle" giving a placement error.

Are there other strategies I should explore? I have started looking at image registration, but it looks like it may have trouble with the variation in the "bump" appearance.

A sample image is attached.
The similarity of the shape suggests trying a correlation/match filtering approach. Specifically, take an image of a single stud bump and match to the full image. Your correlation peaks will be centered around the locations of the bumps in the image. See an example of this in the Image Processing Tutorial. You may have to experiment with different "generic" stud bump images until you find an ideal template.

WaveMetrics, Inc.