Does the work need some skills in Fourier Analysis, as I guess that is needed
for some image treatment? And the second questions is if it needs to be written
in Java? If it needs computation time, C++ should be about a 100 times faster.

This is some really interesting work Bruce. Please post more pics of other leafs when you get the chance. I too see the veins but they are in positions different to where they should be-very strange stuff so it will be interesting to see if other images also have this same effect.

Bruce, have you made any further advances with your leaf photos? Awaiting more of these photos as the previous were very interesting.

Pascuser wrote on Jun 21^st, 2012 at 9:51am:

Pascuser,

I am happy to hear from someone interested in experimenting with this fascinating technology and I hope more will join us.  Please feel free to ask more questions and I'll do my best to answer. 

Your are correct about using only one picture, but in my early work I computed  the intensity relative to an arbitrary, constant pixel value, sometimes using a value of 256.  In order for this to work it is assumed that:
1. The white background light intensity is perfectly uniform or very nearly perfect.
2. AND that none of the pixel values in the phantom area of the photo exceed 255.

The method I first used to obtain the phantom leaf images used a single photo.  With this method the actual pixel value in the missing leaf area of the photo was subtracted from an arbitrary, fixed value (often 256) to yield a new value.  This new valve was then used  in pseudo coloring to bring out the phantom leaf structure in the image. 

For example for an 8 bit per channel, RGB (standard JPG file format.) image I chose 256 as the arbitrary  value.  My filter algorithm was then:
New Value = 256 - Actual Pixel Value

The New Value was then used to select a pseudo color from a crude Lookup Table to generate the final image.  It took quite a bit of experimenting with pseudo color selections to make the phantom leaf structure clearly visible.  The picture at the bottom of the page at http://afterlife-knowledge.com/phanpics.html was one of the better of these pseudo color selections.  The vein structure turned out to be dark relative to the surrounding areas.

It is important to understand how critical it is that assumption number 1 above be true in order to obtain phantom leaf structure images.  Ideally, if you took a photo of just the white background, without any leaf, the intensity of EVERY pixel in that image would be EXACTLY the same value as all other pixels of that pixel color.  In other words in an RGB image every red pixel would be the same value as all other red pixels.  Green pixel values can  be different from the red or blue RGB image values, but all green pixels would, ideally be the same value.  Same goes for the blue pixel values in an RGB image of just the white background without any leaf.    

To see how uniform my white background was in my early work I used my filter algorithm to generate a pseudo colored image for each of the RGB pixel colors.  In this way I could verify that my photos later taken with the leaf met that first assumption above.  Then, since the background is uniform intensity I could use an arbitrary value (like 256) in the pseudo coloring of the phantom images.  In those images I was seeing the differences in pixel intensity values that the phantom leaf structure imposed over the previously uniform background intensity.

In practice I found it impossible to create an absolutely perfect, uniform intensity background.  But I didn't start seeing the phantom leaf structures in the pseudo colored images until the white background was pretty close to perfectly uniform.  Probably in the order of (I am guess estimating here) a total range of 5 (out of 256).  The more you can do to make your white background pixel intensity (in an image without the leaf as a test photo)  perfectly uniform the better your chances of getting good phantom leaf photos.

I hope that answers your questions. If not please feel free to ask more.

In my present work I am trying a different approach that uses three photos, but I recommend that experimenters start with the single photo method as it helps you learn about other factors.  For example, how do you chose a leaf?  How soon must the picture be taken after picking?  Depth of field settings?  Illumination light frequencies?

The three photo method is an attempt eliminate the need for a near perfectly uniform background.  The camera must not move AT ALL when taking Photo#2 and #3 using this three photo method.  Use a cable release, or ideally, a remote electronic trigger to trigger the shutter.  If you use your finger to push the camera shutter button the camera will move enough to make this method not work!  Also you MUST lock the aperture, shutter speed and focus to the same place for all three pictures.

Photo #1.
A photo is first taken of the original, intact leaf without a portion removed.  This is a reference photo used later to visually compare the actual physical leaf structure to any phantom leaf structure obtained from the next two photos.  The leaf in this photo must be positioned in the same location (relative to the camera and white background) as it will be in the next photo, that is, Photo #2.  Focus on the leaf, lock that focus, select an appropriate aperture and shutter speed and lock those so they are the same for all three photos.  This method will not work if you fail to do this!

Photo #2
Cut way a portion of the leaf quickly and carefully.  Then take photo #2 without moving the leaf from its location relative to the camera and background.

Photo #3
Remove the entire leaf from its location so that it will not be seen in this photo.  Drop it on the ground is the easiest.  Then quickly take photo #3.  This is an image of just the background.

The algorithm for pseudo coloring in this method obtains the New Value by subtracting the pixel values of Photo#2 from Photo#3.  Since the actual background is the same in both images this subtraction removes the background and any non-uniformity in that background.  It gives a result that is only the background difference induced by the phantom leaf structure.  The background doesn't need to be perfectly uniform because this method subtracts the background from the resulting image.  All that is left to do is to pseudo color that resulting image and visually compare it to Photo#1.

What is taking me so long at this point is that I am trying to do this using 16 bit per channel RAW image files (converted to TIF files) instead of 8 bit per channel JPG files for greater resolution.  There are camera sensor noise issues to deal with as well as writing filter programs to handle these high resolution RAW or TIF images.  I would suggest experimenters use JPG format images (perhaps super fine resolution JPG) if you attempt this method.  Also be careful that your subtraction of the two images does not result in a negative number! Those negative number results are automatically set to zero making them useless.  Little details like that can prevent obtaining phantom leaf images.  Details, details.

Well, like I said at the beginning, I hope this helps answer your questions and I hope more experimenters will join in.

Until there are more questions, or I have some great results to share . . .

Good luck!!

Bruce

ottawa1 wrote on May 9^th, 2012 at 3:28pm:

ottawa1,

I have seen the shift of vein structure locations in other phantom images, it is actually pretty common.  Often some or most of the veins are in, or close to, their physical leaf positions with only a few showing large displacements.  I don't know what causes these shifts.  Maybe the "fabric of empty space" they reside in is flapping in some unseen "empty space breeze"   ?

I wish I had the technical horsepower to use videos instead of single images so I could observe this shift over time.   That might yield some really interesting results.

Bruce

I have things to show you, but the system tells me that I can't post pictures and links if I have not more than 3 posts in the forums.

So this post is the third one, I hope next one will be more interesting.

I designed a small program in Delphi to do all the adequate manipulations. It allows me to load the picture, and to do what Bruce did:

I decide a threshold value for pixel green intensity that allow to detect the background of the hole in the leaf. For every pixel with grenn intensity above this threshold (the background pixels, white approximately) I make a substraction.

I substracted the threshold to the pixel value:

pixel[i,j] green intensity - threshold -> result value green intensity
and the same for blue and gred intensity for the same pixels.

Then I used these result values to colorize the pixels with a random palette I first generated. How? I completed a table with random value between 0 and 255, with three lines: one line of random values for R, one line for G and one line for B.

Then when the result value for a pixel in green intensity is 5 for sample, I take the 5th value on the table in the G line and that defines the new intensity for this pixel.

I have designed a button to click to generate a new random palette of coulours (a new random table) to have another colorizing. Then I can change both the random palette and the threshold value. I take a picture screen capture whrn I have something interesting.

That is what I did.

Now the results:

First, the same pictures I posted in last message computed with this program:




A new experiment done today with another leaf (coming from another tree):

Before I cut a hole in the leaf:


For this test I used my Numeric picture camera and I did three pictures: one with the leaf in "good health", one with the same leaf amputed and one without the leaf to have the backgound, picture camera being at the exact same place. But my picture camera changes aperture when it decides and I can't prevent it because it is all automatic (a cheap one). So the background intensity is not the same with the meaf and without (because without leaf, there is a lot of white intensity and the camera automatically changes the aperture to reduce the luminosity).

I got this result, when computing the picture result of the substraction of background picture from the amputed leaf picture:


Using the picture taken directly, with no background substraction, and using my program I got (in gray levels here for this test):


Now a new experiment with my webcam, another leaf from the same tree. It has the advantage not to have autofocus or else: all is parametrized with software (IRIS here).

I don't know why I did not posted here a picture with background substraction, I only captured the work I did with direct use of my program, with no other background picture used:

In gray levels:


With color, another capture and computing of the picture: