Image processing questions

[tkr]

Hello,

I had a couple of questions on image processing

a) Any recommendations on a good book to learn about (practical) image processing - calibration, rotation, etc? I'm an electrical/computer engineer so the math does not bother me. Just never dabbled in that side of EE.

b) is there an easy way to tell the distance of the book from the camera ? (perhaps using the calibration picture and image processing ?) Would want to use this to compensate for the change in focus distance as the pages of the book are turned.

c) Lets say I'm using a 90 degree platen and trying to photograph both pages at the same time, ,and then using s/w to de-keystone the individual pages. In terms of image clarity, is there a cost to doing that than if I were to have a photograph taken of each page, with the camera positioned normal to the page surface ?

d) During key-stoning, I assume that one computes a transform to apply to the image. Does this mean that somewhere in the process, one is able to find out what the roll/pitch/yaw/ is of the camera with respect to the book surface ? Intended use is for that information, is that it would help with adjusting the camera tripod, so that the book surface is normal to the camera - right now I'm doing it by trial and error, since my tripod does not have any 'angle' adjustments. (

Thanks
TKR

[og200]

Hi tkr,

These are some really interesting questions. I'm no expert but here's my 2c worth.

Re. c) Pelle posted a video of an app that does what you're after (video is here and the thread is here); I don't have experience with the quality (though it would be easy enough to do a quick test with hand correction and see how it looks). The software looks like it does marker detection on the image and uses the marker coordinates to split and deksytone the page. (There are lots of libs that do marker detction; here's a video example in OpenCV.) It would be great to have this sort of software in the community (if it isn't already).

Re. d) the keystone correction I do in ScanManager is quite simple -- it's just a perspective warp from one quad to another. OpenCV takes care of finding the chessboard, I just tell it to transform the points on the chessboard to new points that lie on a nice rectangle of the right proportions. It will then calculate a 3x3 transformation matrix to do that warp (see getPerspectiveTransform). If you wanted to extract rotation/translation components you could do a matrix decomposition on the transformation matrix.

[rob]

On (a) and possibly (b) (I haven't looked at this in a while), Learning OpenCV: Computer Vision with the OpenCV Library is an excellent book which teaches you the basics, plus how to use the open source OpenCV library to do the image processing.