histogram functions in plantcv

[DannieSheng]

Two types of histogram functions are used in plantcv: cv2.calcHist is used in pcv.analyze_color and np.histogram is used in pcv.visualize.histogram.

cv2.calcHist(images, channels, mask, bins, ranges): this function is capable of taking a variable "mask" and calculate histogram only on the mask area.

np.histogram(a, bins, range): this function takes an array as an input and calculate histogram. So if we want to calculate histogram only based on masked area (that's probably most the cases we use the function), we will have to apply the mask first and calculate the histogram based on the masked image. (this is the case in the function pcv.visualize.histogram.

Only when set bins to be 255 and range to be [1,256] can these two method have the same result.
hist_data = [int(i[0]) for i in cv2.calcHist([img], [0], mask, [255], [1, 256])]
hist_data, hist_bins = np.histogram(masked, 255, (1, 256))

Which means we exclude value 0 when calculating histogram.
That is because when using the histogram method in np, currently we are calculating histogram on masked image, not the masked part of the image, i.e. a lot of 0s.

Though it might be fine for most cases, but to be more general I think it makes much sense to include 0, or at least capable of including 0.

Since cv2.calcHist is capable o f calculating histogram with or without mask, I propose to use cv2.calcHist in plantcv.

[DannieSheng]

Used wavelength -> web color code function to visualize histogram for RGB image.
Application on hyperspectral image still testing.

[DannieSheng]

I just realized that this wavelength to hex color might my unnecessary at this point: the colors for plots will be hard coded for RGB images. For hyperspectral images we currently only allow input as an nd-array, so there will be no corresponding wavelength information.

[DannieSheng]

Just got the testing result for hyperspectral image. It seems that if showing all bands, we should not show the legend... It seems not necessary and not suitable that we show histograms of all bands by default.

[DannieSheng]

A pull request will be created for functions pcv.visualize.histogram and pcv.analyze_color, where pcv.analyze_color make histograms by calling pcv.visualize.histogram.
For pcv.visualize.histogram:

1. Uses np.histogram
2. Takes nd-array as input image (RGB image, grayscale image, thermal image, hyperspectral image?)
3. No options to change color

[DannieSheng]

Solution present in #698

[DannieSheng]

There are 5 histogram related functions:
(1)pcv.analyze_color
(2)pcv.analyze_nir_intensity
(3)pcv.analyze_thermal_values
(4)pcv.hyperspectral.analyze_spectral
(5)pcv.hyperspectral.analyze_index
There is an input histplot for all functions (2), (3), (4) and (5) such that when set to True, a histogram will be plot. (Default value: False)
Currently functions (1) (2) (3) have already been updated based on the updated pcv.visualize.histogram in pull request #698 . In the meantime, there are several discussions I would like to bring up.

For function (1) there is a little different: the input is called hist_plot_type and the default value is None, other acceptable inputs are "rgb", "lab", "hsv" and "all". The hue stats would be saved no matter what value is set for hist_plot_type.

For function (4) there is also a little bit different situation. Although the variable says "histplot", what actually plotted is the average spectra of the pixels from the masked area, not a "normal" histogram.

Based on the information above, there are several proposals:

1. Starting printing warning messages regarding deprecating histplot for function pcv.hyperspectral.analyze_spectral (4)
2. Add another function in visualize sub-package called histogram_hyper (suggestions of a better function name welcomed!!!) that takes a hyperspectral image, a mask, as well as a list of wavelengths and plots the histograms for input wavelengths. The color of plots would be decided by the desired wavelengths.
   pcv.visualize.histogram_hyper(array, mask, wvlengths=[480,550,670])
   Sample result:

[nfahlgren]

Today's discussion

• Deprecate histplot inputs in analyze_nir_intensity, analyze_thermal_values, analyze_spectral, and analyze_index
  • Leave histplot for now and print deprecation warnings
• Create deprecation function
  • Prints deprecation warnings by pretending "Deprecation warning:" to message
  • Print current version of PlantCV
• Deprecate hist_plot_type in analyze_color
  • Replace with a colorspaces input
  • Print deprecation warning for now

[DannieSheng]

For added histogram_hyper function

• print out a warning message when desired wavelength is "too far" from available wavelengths (2x the spectral resolution)
• print out a warning message when more than one desired wavelengths is smaller than the smallest available wavelength or larger than largest available wavelength

[DannieSheng]

Updates on 03/23/2021:
For visualize/histogram, add a flag indicating whether or not return histogram data, default value is False, such that users can continue to use this function normally, with only histogram plot returned.