embedart: write candidate image only if it is "similar" to already embedded one (suite)

[Kraymer]

Original issue : #848
Continuation of work started by PR #966

Taken from the docs :

The compare_threshold option defines how similar must candidate art be regarding to embedded art to be written to the file.
This works by computing the perceptual hashes (PHASH_) of the two images and checking that the difference between the two does not exceed a given threshold.
The threshold used is given by the compare_threshold option:

• use '0' to always embed image (disable similarity check)
• use any positive integer to define a similarity threshold. The smaller the value, the more similar the images must be. A value in the range [10,100] is recommended.

Additional commit infos :

• 26ec2b8: I kinda hacked ArtResizer original goal, using it only to inform me if IM is installed or not. In the long term we might want to do some other basics operations and renaming the class into ArtProcessor or something would fit better I think.
• a06c27 : what's tricky here is that compare return code is 1 in case of non match, yet the two images in such case may be judged as similar from a human point of view. So I needed to interpret the printed diff value even when CalledProcessError is raised. Maybe this edit to command_output could be done more elegantly ?
• 187497c: I don't know what range of values can compare -metric PHASH return ([0, ?]), so I decided to let user set the threshold and hinting at a reasonable max in the docs (100)

[Kraymer]

Problem with python 2.6:

  File "/home/travis/build/sampsyo/beets/beets/util/__init__.py", line 643, in command_output
    raise subprocess.CalledProcessError(proc.returncode, cmd, stderr)
TypeError: __init__() takes exactly 3 arguments (4 given)

If someone have an idea how to work around that... by raising a custom error (matching the py2.7 error signature) that we will define at the top of the file?

[sampsyo]

Hmm... it looks like you're using this to capture the stderr of the command when the invocation fails. You probably already thought about this, but is this really necessary? Could you invoke the comparison command so that, when it fails, you know the images are too dissimilar?

Alternatively, you could consider just not using the command_output utility here; other uses of the utility don't care about the stderr so the use case is different enough to justify reimplementation.

[Kraymer]

Could you invoke the comparison command so that, when it fails, you know the images are too dissimilar?

I thought not, but after checking compare --help it seems setting -dissimilarity-threshold when calling could help.

Edit (after testing) : actually, no. Changing the dissimilarity-threshold does not yield any result regarding the return code value.

[Kraymer]

@sampsyo if it s OK with u, I d like to merge that PR. Got another one comin up =]

[sampsyo]

LGTM! All merged up. See minor changes above (mainly docs, some style).

[Kraymer]

added a note concerning the doc.
And yeah thanks to get rid of all these passive forms I use when I write.