Flatbed scanner as camera.

General spoken, vision systems use a camera, optics, illumination and a computer with vision software. In some situations when high resolution is necessary a vision designer can choose between a high resolution camera, more lower resolution cameras and create one large high resolution image or one scanning camera system. In the last situation a vision designer can choose between a line array camera and a scan in one direction or a 2D camera and a two direction scan. Anyway in some situation an expensive solution.
In most cases there is no alternative.
But..... in a few cases a flatbed scanner can be used to acquire a high resolution image. Image red dots small JAIMS
J.A.I.M.S. used in that particular situation a flatbed scanner for inspection of a 'Wafer' filled with about 3000 glass bolls width a size of about 0,5mm.
Each of these glass bolls are fixed in a holder and separated from each other. (see the image of the glass balls and holders) The pitch is about 3,0 mm and the size of the wafer diameter (field of view) is 180mm. The quality of these balls are inspected manually (using a microscope) on both sides and the rejected ball is marked with a red or blue marker. One side in red the other side in blue.
These matrix of balls are 'glued' on a silicon based foil and are diced by a sawing process.
Further on in the handling process a pick and place robot will pick the accepted ball (ball without a red of blue marker) out of the wafer and place it on the required assembly position. Since the pick and place robot doesn't have an extra camera on the gripper it requires a mapping file of the accepted and rejected balls, so it will get the accepted balls out of the wafer.
Because of the manually inspection it is difficult and infeasible to create an error free mapping file manually. In this situation an automatic mapping file generator built around a flatbed scanner is a good solution.

Realize the required resolution, size (diameter) of the ball is 0,5mm. A flatbed scanner resolution is selectable in 300dpi, 600 dpi, until 3600dpi (or more). It depends of the flatbed scanner if the highest resolution can be used for whole scanner size.
In this situation we needed a scan size (field of view) of about 200mm. The size of the wafer is 180mm so +/- 10mm can be used to position the wafer on the scanner.

J.A.I.M.S. used the 600dpi setting, the resolution was enough to do the job and the scanner is not getting really slow. Also the size of the acquired image is within certain limits.
Resolution: 600dpi is about 42micron per pixel. Ball diameter 500micron results in a 500/42= 11,9 pixels (say 11 pixels one direction). Total amount of pixels is about 90.
Of course a lot of border pixels are not used and the marker is not covering the whole ball. But more than enough pixels for color detection.
To scan an area of 200mm square using 42 microns per pixel, the image is 4800x4800pixels or 23Mpixel (say 25Mpixel three colors). Realize these images are color images.

Hardware Set up

This mapping file generator is built around the flatbed scanner.
A large back light (white) is positioned on top of the flatbed scanner, so light is getting through the ball in the direction of the line array camera in the scanner. Scanner slit small JAIMS
Both sides are detectable by using this back light. The glass balls are 'glued' on a silicone wafer and therefor the wafer is not totally flat.
To prevent the balls of any extra unwanted scratches the waver is positioned severals millimeters above the glass plate. The construction is comparable with a slit of the letterbox.

Software

The software is really complicated for this inspection so a few important points are given here. The flatbed scanner does have a line array camera which moves in one direction. In the direction it moves the camera image is perpendicular to the scan surface, in the other direction the camera is not perpendicular to the surface.
Detected pattern 01 JAIMS Detected pattern JAIMS Marker positions 01 JAIMS So the scan created using this scanner is in two directions different. Fortunately the focus depth is sufficient enough.
In this particular situation the wafer is not flat it does it sags a little bit.
On the wafer are several alignment marks so the software can align the wafer with respect to the image. The software also compensates this sag of the wafer using a fourth degree polynomial fit.
Unfortunately this was not enough so an extra template match algorithm is used to detect each ball accurately within one pixel!. Here two measure results are given (test patterns), green is oke, red en blue are rejected balls.
Also the positions of the markers are given. Yellow is no ball and black is outside the testregion.
Front panel 50perc Screen print  and settings 50perc JAIMS
The light intensity is in one direction less homogeneous than in the other direction. Therefore the intensity over the whole image is compensated using the three color image. After all these correction a differential color detection algorithm is develop to the very high inspection quality level.
The vision system does have a lot of extra features e.g. change software settings, save (color) images, change settings wafer, wafer type selection, tests on saved saved images, debugging. Here two plots of the user panel and the panel with a lot of extra information.

Remarks: The possibilities of a flatbed scanner in the 'vision world' is limited but it is a cheap high resolution sensor.
Compare the alternative, a 25Mpix camera and a suitable lens. The software is basically the same only a camera is faster.