Histogram of LCH image

[schmijos]

I'm trying to get an ndim histogram of the CH part of an image:

(here's the image: )

img = Vips::Image.new_from_file('spec/fixtures/files/sample.png', access: :sequential).colourspace(:lch)

img[1..2].to_a
=> [[[102.3130874633789, 39.18145751953125],
  [101.67424774169922, 39.16696548461914],
  [0.4031951427459717, 110.17615509033203],
  [118.2603988647461, 135.60243225097656]],
 [[102.3130874633789, 39.18145751953125],
  [102.67890930175781, 39.159664154052734],
  [118.54254913330078, 135.6748809814453],
  [0.00013431724801193923, 326.3099365234375]],
 [[133.49240112304688, 306.25054931640625],
  [0.33749398589134216, 201.97052001953125],
  [115.21880340576172, 328.395263671875],
  [95.71614074707031, 102.8958511352539]],
 [[0.019027704373002052, 326.1108093261719],
  [132.93775939941406, 306.2183532714844],
  [95.6482925415039, 102.39655303955078],
  [96.01411437988281, 102.91535186767578]]]

img[1..2].hist_find_ndim(bins: 5).to_a
=> [[[16], [0], [0], [0], [0]],
 [[0], [0], [0], [0], [0]],
 [[0], [0], [0], [0], [0]],
 [[0], [0], [0], [0], [0]],
 [[0], [0], [0], [0], [0]]]

Why are all values counted into 0,0,0?

The documentation says something about casting (hist_find_ndim), but still most of the values should be in range, shouldn't they?

[jcupitt]

Hi @schmijos,

You are taking the histogram of a float image, so it's casting it to uint16 before scanning. Five bins on a range of 0 - 65535 means each bin is more than 10,000 wide, so all your values fall in bin 0. If you cast to uint8 (uchar) yourself, you should see something more expected.

It's best to fix the range, of course. C is roughly -128 to 128, and h is 0 - 360, so you can move them to 0 - 255 like this:

#!/usr/bin/ruby

require "vips"

x = Vips::Image.new_from_file ARGV[0]
x = x.colourspace("lch")[1..2]
x = (x * [1, 256 / 360] + [128, 0]).cast("uchar")
arr = x.hist_find_ndim(bins: 5).to_a
puts "histogram: #{arr}"

I see:

$ ./schmijos.rb ~/pics/tiny.png
histogram: [[[0], [0], [4], [0], [12]], [[0], [0], [0], [0], [0]], [[0], [0], [0], [0], [0]], [[0], [0], [0], [0], [0]], [[0], [0], [0], [0], [0]]]

[jcupitt]

(perhaps the histogram functions should always cast to uchar, with ushort as an option you have to turn on explcitly? It might be less surprising ... though it's too late to change the behaviour now, unfortunately)

[jcupitt]

I've tried to clarify the doc comments.