The figure below shows what can be done with this package. The user can modify the appearance of objects in photographs. For example, they can make human skin smoother or, conversely, make blemishes more visible. They can also enhance the gloss of fruit or make it look wilted. Based on image processing techniques, the user can automatically perform such photo editing effects by simply calling on a single R function.
Details of this package (background, algorithm, usage, etc.) is described in the article below:
Tsuda and Kawabata (2023). materialmodifier: An R package of photo editing effects for material perception research. Behavior Research Methods. https://link.springer.com/article/10.3758/s13428-023-02116-2
If you cite this R package, please also cite the following paper (because this technique is an R implementation of the algorithm proposed in the paper below).
Boyadzhiev, I., Bala, K., Paris, S., & Adelson, E. (2015). Band-sifting decomposition for image-based material editing. ACM Transactions on Graphics, 34(5), 1–16. https://doi.org/10.1145/2809796
Mac users need to install XQuartz (https://www.xquartz.org/).
There are two ways to install the package: 1. via GitHub, 2. via CRAN.
(The reason for recommending installation via GitHub is that if a version of the package is updated by adding new features or removing bugs, installing from GitHub will allow you to use that latest version. There can be a time lag of several weeks before CRAN data is updated.)
You can install the development version of the
materialmodifier package via GitHub, by using the
devtools package.
# install the devtools package
install.packages("devtools")NOTE:
To install a package from GitHub,
- On Windows,
Rtools
needs to be installed.
- On Mac, in some environments, installation of XCode may be needed.
After you have installed the required software stated above, install the package as follows:
# install the materialmodifier package
devtools::install_github("tsuda16k/materialmodifier")Then, attach the package.
library(materialmodifier)(I do not recommend installing via CRAN because the package available on CRAN is not the latest version and lacks certain features. Although the package version will be updated soon, I suggest installing it via GitHub in the meantime.)
The package is available on CRAN. It can be installed with:
install.packages("materialmodifier")Then, attach the package.
library(materialmodifier)The materialmodifier package contains data for an image,
which is useful when you want to quickly try out a material editing
effect. The variable name for this image is face.
Internally, this data is a numeric array with the size 500 x 500 x 3 [y-coordinate, x-coordinate, color channel], which means a height of 500 pix, a width of 500 pix, and three color channels (Red, Green, and Blue). Each element of the array represents a pixel value, which can be between 0 and 1.
dim(face)
#> [1] 500 500 3The image size information can be displayed by typing the variable name.
face
#> image: 500 [height] x 500 [width] x 3 [colour channels]To plot an image, use the plot() function.
plot(face)
To load an image in your computer, use the im_load()
function.
im = im_load("path/to/your/image.jpg")
plot(im)The jpg, png, and bmp formats are supported.
You can load an image from the web (URL). For example,
im = im_load("https://raw.githubusercontent.com/tsuda16k/materialmodifier/master/notes/meat.png")
plot(im)will load an image of roasted meat.
In the following example, we use the face image contained in the
package.
For consistency purposes, the face image is labeled as
im.
im = faceUse the modif() function to apply a material editing
effect to an image.
(Note: the function name is modif, not
modify)
Below is an example of the shine effect.
# apply the shine effect
im2 = modif(im, effect = "shine", strength = 3.0) # may take some seconds
plot(im2) # see the result
On the left is the input image, and on the right is the result of the
shine effect.
The strength parameter is used to control the strength
of the image processing effect. A larger value of strength
will produce a stronger effect:
The strength parameters, from left to right, are 0.1, 1,
2, 3, and 4.
Note that when the strength parameter is 1, the
modif() function just returns the input image. Therefore
the second image from the left is the same as the input image.
When the strength parameter is between 0 and 1, the
output image becomes less shiny than the original (see the leftmost
image above).
Below is an example of the aging effect.
# load the face image
im = im_load("https://raw.githubusercontent.com/tsuda16k/materialmodifier/master/notes/face.jpg")
# apply the aging effect
im2 = modif(im, "aging", 0.1)
im3 = modif(im, "aging", 2.5)
# show the results
plot(im2)
plot(im)
plot(im3)
Compared to the original (center), skin blemishes are reduced on the left and boosted on the right.
The figure below summarizes outputs of the shine and aging effects.
Using the aging effect as an example, the effect of the strength parameter is examined in more detail in the figure below. If the value of the strength parameter is greater than 1, a boosting effect that increases the stains/blemishes occurs; if it is less than 1, a reducing effect that decreases the stains/blemishes occurs.
To get a boosting effect, a strength value of 2 to 6 usually yields reasonable results. The strength parameter can be a negative value, but in most cases setting a negative value will produce unrealistic results (e.g., contrast reversal)
This package has several other effects in addition to the shine and aging effects. All available effects are: shine, spots, rough, stain, blemish, shadow, aging. A visual summary of these effects is shown in the figure below.
The first column of the figure shows the name of each effect, and the second column shows the perceptual features controlled by that effect. The third and subsequent columns show the input and output images.
The figure contains a column called BS feature. The BS feature is an important term related to the image processing algorithm (briefly, the image area to be manipulated, extracted from the input image based on a certain criterion). The algorithm achieves image editing effects by decreasing or increasing the weight of the BS feature in the input image.
Since terms such as shine and aging are
aliases for BS features, each effect can also be applied using the name
of the BS features as follows.
# im1 and im2 are identical
im1 = modif(face, "shine", 3)
im2 = modif(face, "HHP", 3)
# im3 and im4 are identical
im3 = modif(face, "aging", 3)
im4 = modif(face, c("HLA", "HHN"), 3)Note that the aging effect is an effect that edits two
BS features, HLA and HHN, so two BS feature names are specified
above.
To better understand the nature of each image editing effect, it is helpful to compare the results of all editing effects on a single image. The figure below summarizes the results of the editing effects on a face and a food image (note that the image in the row with a strength value of 1 is the input image). By comparing the images in the rows with large values of the strength parameter, it is easier to see the characteristics of each effect.
Setting a negative value for the strength parameter often results in an unnatural image (see images in the bottom rows of the figure), but using a large negative value for the strength parameter makes it easier to compare which areas are affected by each effect. For example, the rough effect and the blemish effect produce similar results, but if you compare the images in the row with a strength value of -5, you can clearly see that they are not identical.
Technically, the blemish effect is equivalent to giving both the rough and stain effects at the same time. In order to get a more formal understanding of these properties, we need to know more about the specifics of how image processing algorithms work. See our paper for details.
You can also perform multiple editing effects at the same time. For example, you can apply the shine effect and the aging effect at the same time as follows.
# Applying multiple effects at the same time
im2 = modif(im, effect = c("shine", "aging"), strength = c(0.2, 3))
plot(im2)
The above command simultaneously applies a shine effect of strength = 0.2 and an aging effect of strength = 3, resulting in a less shiny and more blemished image.
Applying each effect in sequence produces almost identical results.
# Applying multiple effects at the same time
im2 = modif(im, effect = c("shine", "aging"), strength = c(0.2, 3))
# Applying effects in sequence
im3 = modif(im, effect = "shine", strength = 0.2)
im3 = modif(im3, effect = "aging", strength = 3)
# pixel-wise mean squared error
mean((im2-im3)^2)
# this value is very small, meaning that both images are almost identicalAlthough you can get the same result by applying each effect in turn (e.g., applying an aging effect to the output of a shine effect), we recommend doing them in a single line, because it saves time needed for image processing.
By using a mask image, you can edit only certain areas within an image, rather than the entire image. To use this feature, you need to prepare a mask image of the same size as the input image you wish to edit.
The mask image is an image in which the area to be edited is white and the rest of the image is black. The mask image does not have to be a binary image; gray can be present (the intensity of the gray will be used to alpha blend the input image with the edited image).
For example, the mask image representing the skin region of the face image in this package is as follows (Image on the right).
Specify a mask image for the mask argument of the
modif function.
# load images
im = im_load("https://raw.githubusercontent.com/tsuda16k/materialmodifier/master/notes/face.jpg")
mask = im_load("https://raw.githubusercontent.com/tsuda16k/materialmodifier/master/notes/mask.jpg")
# apply an effect
im2 = modif(im, "HHP", 3) # no masking
im3 = modif(im, "HHP", 3, mask) # with a mask image
# see the results
plot(im)
plot(im2)
plot(im3)
Left:
input image, Center: edited without mask, Right: edited with mask
An example of the HHP(shine) effect applied to the face image is shown above. When editing without using a mask, the entire image is edited, so the hair is shined as well as the skin. On the other hand, when editing with a mask image that specifies the skin area, only the skin area is changed, while the other areas remain the same as in the input image.
If you observe the two images on the right in more detail, you will
notice that the appearance of the skin area is slightly different
between the two. This is because by using a mask, only the pixels within
the mask region are used to calculate the image features that should be
edited.
(The image processing algorithm uses the distribution of brightness
values in the input image to extract each BS feature. The distribution
(variance) of brightness values changes as only pixels within the masked
region are included in the calculation, thereby changing the criteria
for high and low amplitude, which in turn changes the BS features.)
The modif2() function allows for a precise control over
which image feature (i.e., BE feature) to manipulate. For example,
boosting the LAA feature (low spatial frequency, all (high
and low) amplitudes, and all (positive and negative) signs) can be
performed as follows.
# modify the LAA feature
im2 = modif2(im, params = list(freq = "L", amp = "A", sign = "A", strength = 2 ))
im2 = modif2(im, params = list(feature = "LAA", strength = 2 )) # equivalent to the aboveDetails about image manipulation features (BS features) are described in our paper.
Here are additional examples for using the modif2()
function.
# shine effect (boost the HHP feature)
shine = list(freq = "H", amp = "H", sign = "P", strength = 2)
plot(modif2(face, params = shine))
# shine effect (equivalent to the above)
shine2 = list(freq = modif_dim(face)$high, amp = "H", sign = "P", strength = 2)
plot(modif2(face, params = shine2))
# you can specify a feature name directly, instead of specifying freq/amp/sign separately
plot( modif2(face, params = list(feature = "HHA", strength = 2)) )
plot( modif2(face, params = list(feature = "1HP", strength = 3)) )
# apply multiple effects at the same time
blemish = list(feature = "HLA", strength = 0.1) # less blemished
smooth = list(feature = "HHN", strength = 0.2) # smoother
plot(modif2(face, params = list(blemish, smooth)))As with the modif function, a mask image can be
specified in the modif2 function.
# apply an effect using a mask image
im2 = modif2(face, params = list(feature = "HHP", strength = 3), mask)The specifications of the modif() and
modif2() functions are summarized below.
The modif() function has 8 arguments.
modif(im, effect, strength, mask = NA, max_size = 1280, log_epsilon = 0.0001, filter_epsilon = 0.01, logspace = TRUE)Table of arguments of the modif() function:
| Argument | Meaning | Value | Default |
|---|---|---|---|
| im | Input image | an image object | |
| effect | Effect name | Either “gloss”, “shine”, “spots”, “blemish”, “rough”, “stain”, “shadow”, or “aging” | |
| strength | Strength of effect | a float value or a float vector | |
| mask | Mask image | an image object | NA |
| max_size | Image resolution limit | an integer | 1280 |
| log_epsilon | Offset for log transformation | a float value | 0.0001 |
| filter_epsilon | Epsilon parameter of the Guided filter | a float value | 0.01 |
| logspace | Log transformation flag | a logical value | TRUE |
The im is an image object we can get by using the
im_load() function.
The effect and strength parameters have been
described above.
The mask is an image object we can get by using the
im_load() function.
The max_size parameter can be used to restrict the image
resolution. If the shorter side of the input image is larger than this
value (the default is 1280), input image is resized before applying
effects. For example, when the input image has 1024 x 2048 px
resolution, and if max_size is 512, then the input image is
first resized to 512 x 1024 px. Because the modif() function is very
slow for large-resolution images, it is useful to limit the image
resolution to speed up the image processing.
The log_epsilon and filter_epsilon are
parameters that are used for image processing procedures. You need not
change this value in most cases.
The logspace parameter is TRUE by default, meaning that
image processing is performed after log transforming the L component (L
in Lab space) of the input image. If FALSE, no log transformation is
performed. Without log transformation, the output image is prone to edge
artifacts (areas of edges in the image tend to look unnatural), so that
it is recommended to use the default setting.
The modif2() function has 7 arguments.
modif(im, params, mask = NA, max_size = 1280, log_epsilon = 0.0001, filter_epsilon = 0.01, logspace = TRUE)Table of arguments of the modif2() function:
| Argument | Meaning | Value | Default |
|---|---|---|---|
| im | Input image | an image object | |
| params | A list of parameter values | a list | |
| mask | Mask image | an image object | NA |
| max_size | Image resolution limit | an integer | 1280 |
| log_epsilon | Offset for log transformation | a float value | 0.0001 |
| filter_epsilon | Epsilon parameter of the Guided filter | a float value | 0.01 |
| logspace | Log transformation flag | a logical value | TRUE |
For the params parameter, set a list of material editing
parameters:
# shine effect (boost the HHP feature)
shine = list(freq = "H", amp = "H", sign = "P", strength = 2)
plot(modif2(face, params = shine))
# shine effect (equivalent to the above)
shine2 = list(freq = modif_dim(face)$high, amp = "H", sign = "P", strength = 2)
plot(modif2(face, params = shine2))
# you can specify a feature name directly, instead of specifying freq/amp/sign separately
plot( modif2(face, params = list(feature = "HHA", strength = 2)) )
plot( modif2(face, params = list(feature = "1HP", strength = 3)) )
# apply multiple effects at the same time
blemish = list(feature = "HLA", strength = 0.1) # less blemished
smooth = list(feature = "HHN", strength = 0.2) # smoother
plot(modif2(face, params = list(blemish, smooth)))The other parameters are the same as the modif()
function.
The get_BS_energy() function can be used to calculate
the energy of each BS feature in an image. The energy of a BS feature is
defined as the sum of its squared luminance: All the pixel values of a
specific BS feature image are squared and then summed.
get_BS_energy(face)
#> feature energy normalized
#> 1 HHP 5.131705e-04 0.034824299
#> 2 HHN 7.377783e-04 0.050066426
#> 3 HLP 8.040328e-05 0.005456253
#> 4 HLN 5.821145e-05 0.003950291
#> 5 LHP 3.514644e-03 0.238507480
#> 6 LHN 7.169572e-03 0.486534868
#> 7 LLP 1.570909e-03 0.106603543
#> 8 LLN 1.091301e-03 0.074056840
#> 9 total 1.473599e-02 1.000000000The output of the get_BS_energy() function is a data
frame, which has the energy values of eight BS features (rows 1-8) and
the total energy (sum of the above eight, row 9). The
normalized column is obtained by dividing each energy value
by the total energy value.
By specifying a mask image, only specific areas in the image can be targeted for calculation.
mask = im_load("https://raw.githubusercontent.com/tsuda16k/materialmodifier/master/notes/mask.jpg")
get_BS_energy(face, mask)
#> feature energy normalized
#> 1 HHP 7.320857e-05 0.007683400
#> 2 HHN 2.737560e-04 0.028731286
#> 3 HLP 7.994283e-05 0.008390174
#> 4 HLN 4.045785e-05 0.004246140
#> 5 LHP 3.738960e-03 0.392412074
#> 6 LHN 2.059903e-03 0.216191301
#> 7 LLP 2.607734e-03 0.273687387
#> 8 LLN 6.541859e-04 0.068658238
#> 9 total 9.528148e-03 1.000000000This is useful, for example, if you want to analyze only the skin area of a face image.