Understanding histogram() in Pillow

4
votes

From the docs:

im.histogram() => list

Returns a histogram for the image. The histogram is returned as a list of pixel counts, one for each pixel value in the source image. If the image has more than one band, the histograms for all bands are concatenated (for example, the histogram for an “RGB” image contains 768 values).

I understand that there are 256 values for Red, 256 for Green and 256 for Blue (256 * 3 = 768).

for i, value in enumerate(im.histogram()):
    print i, value

Produces:

0 329
1 145
... (skipping some)
256 460
... (skipping some)
767 3953

My question is: does that mean that there were:
329 pixels that had a value of R = 0, G = 0, B = 0 and
145 pixels that had a value of R = 1, G = 0, B = 0 and
460 pixels that had a value of R = 256, G = 1, B = 0 and
3953 pixels that had a value of R = 256, G = 256, B = 256 etc. ?

Is that how I should read the output?

3
pythonhistogrampillow
Given that the documentation says jack about the order, I'd generate 3 separate pics of the primary colors and see what the histogram function outputs. – Peter M
Thanks, that sounds like a good idea. Would still be interesting to see what the order means though. – e h
I'm sure if you wanted to work through the source code you could establish the actual order, but sometimes the simplest methods are the best. BWT when you do figure it out, post your answer back here. – Peter M
@PeterM: Looks like the data is for each channel, see voithos' answer below. – e h

3 Answers

10
votes

I haven't tested, but from the documentation, the wording seems to indicate that the histogram only applies to each channel (e.g. Red, Green, Blue) individually.

If the image has more than one band, the histograms for all bands are concatenated (for example, the histogram for an “RGB” image contains 768 values).

So, no, the examples you gave aren't really correct. 768 values is just 256 * 3, which is the number of possible red values, plus the number of possible green values, plus the number of possible blue values independently. It does not represent all possible combinations of red, green, and blue, which would instead be 256 ^ 3 == 16777216.

From what I can see, the interpretation of your example histogram values should be:

329  pixels with value of R = 0, G = ?, B = ? and
145  pixels with value of R = 1, G = ?, B = ? and
...
460  pixels with value of R = ?, G = 1, B = ? and
...
3953 pixels with value of R = ?, G = ?, B = 256
3
votes

No, you don't know how many pixels had (e.g.) R=0, G=0, B=0. That would require a histogram with something like 16 million entries.

You only know how many had R=0, how many had G=0, and how many had B=0, taken independently.

It would be entirely possible for the R=0 pixels to have a great many different amounts of G and B.

0
votes

if the mode is RGB to open a image, 768 mean is Red grey list[0-255 pixs nums]+Green grey list[0-255 pix nums]+ Blue grey list[0-255 pix numbs]

im is a Image object by RGB

In [59]: x = reduce(lambda x,y:x+y,im.histogram())

In [60]: x Out[60]: 47191725

In [61]: im.size Out[61]: (4731, 3325)

In [62]: 4731*3325 Out[62]: 15730575

In [63]: 4731*3325*3 Out[63]: 47191725

In [64]:

or other way:

r, g, b = im.split()

len(r.histogram()) = 256

len(g.histogram()) = 256

len(b.histogram()) = 256