Scanning the Object in Parts and Image Reconstruction

Dear All,

I am working on a project of Object scanning, Detection and Image classification. Initially, the experiment setup is consist of three components: Light source, Object and CCD camera. The object will be illuminated by light source in parts with spot by spot scanning and every spot will be detected by camera. The spot size around 100*100 pixels, where the camera resolution is 1294*964. The experiment is conducted in dark environment so that only the illuminated spot will be detected and separated from camera's whole pixels. We need to follow this scanning process instead of capturing the whole object in a single shot. It is due to the final application, where we will use other type of detector with limited pixel area and thus by scanning, we can improve the overall resolution. Part of this setup sometime used in Remote sensing, 3D imaging, Airborne laser scanning etc.

Fig.1 shows the original image that is going to be scanned by spots with LED light. Fig. 2 is the image with all the spots as captured and selected by camera. The reconstructed image (fig.2) is not in good shape like original. Due to the circular shape of spot, some portions are dropped and also, I am confuse how to make a good outline of this image. Can anyone give me guidelines to improve the quality of reconstructed image as generated from the scanning process? What are the functions of Image Processing toolbox of MATLAB are useful for this case?

Answer 1

[Bjorn Gustavsson]

To avoid the having non-illuminated black spots you cant move the light-source the full diameter in the horizontal direction, tight-packed circles does not completely fill the plane. A possible solution would be to move the illuminatin by 2*r*cos(30/180*pi) that way you'll get a complete illumination of the entire object, some regions will be illuminated twice. To work around that problem you could make one scan over a perfectly white object, add those images together (After casting to double) to get yourself a white-light reference (it will be ~twice as bright in the regions with overlapping illuminations) then you can repeat the process for your other objects, and use the white-light image to normalize the intensities.

HTH

[pallab choudhury]

Thanks Bjorn Gustavsson for your reply.

The line "...Move the illumination by 2*r*cos(30/180*pi).." is not clear to me. I guess 'r' is the distance from light source to object, isn't it? And also whats about 30 degree?..Is it means that the circle diameter need to vary when it moves from one spot to another horizontally?

However, In my application, the light spot diameter need to be fixed so looking for some other alternatives.....

[Bjorn Gustavsson]

Ah, I meant the horizontal motion of the illuminated circle. In your Fig2.jpg it seems as if you've translated the illuminated circle so that you have no regions illuminated twice, the circles are approximately edge-to-edge, in a hexagonally close-packed lattice where the image coordinate of the illumination-centre is shifted by 2*r. That leaves black regions that you'll never know anything about. My idea was to shift the illumination centre a smaller distance so that no regions are left unilluminated. Something like this should illustrate the idea:

r = 1;
phi360 = linspace(0,2*pi,361);
plot(r*cos(phi360),r*sin(phi360))
axis equal
hold on
plot(r*cos(phi360)+2*cos(30/180*pi),r*sin(phi360))
plot(r*cos(phi360)+4*cos(30/180*pi),r*sin(phi360))
plot(r*cos(phi360)+1*cos(30/180*pi),r*sin(phi360)+1.5) 
plot(r*cos(phi360)+3*cos(30/180*pi),r*sin(phi360)+1.5)

That way the small lense-like regions would be illuminated in 2 images (but that will only reduce measurement noise). If you then image a completely white scene you'll get and add up those images you'll get your flat-field image (of your composed image) to use for normalization of the real images. Then you can proceed with the same sweep over your object.

The "only" thing to think about is to not forget to cast the data format to double since your image addition might lead to overflow, double(uint8(250) + uint8(250)) = 255...

HTH

[pallab choudhury]

Thanks Bjorn Gustavsson for your direction. I got your points and I will focus on it.

However, without go for very details inside of the Image, How can I extract the outer shape pixel values (as like the line in attached figure 'Image_Outline.jpg') from that scanned image so that the object can have better visulization? Can you suggest some MATLAB functions or process?