Given a jpeg, what is the formula to change the exposure of that jpeg by +/-1 stop or as known as 1 EV? I want to simulate this exposure change. Is there a formula/ method to do so?
I can demonstrate that using ImageMagick, which is included in most Linux distros and available for OSX and Windows from here.
First, at the Terminal command line create an image:
convert -size 512x512 gradient:black-yellow gradient.png
Now, the way to effect +1 stop exposure increase is to composite the image with itself using the Screen blending mode - it is available in Photoshop and ImageMagick and is described here.
So, the formula to composite image A with image B is:
1-stop brighter image = 1-(1-A)(1-B)
but as we are compositing the image with itself, A and B are the same, so we effectively have
1-(1-A)(1-A)
ImageMagick refers to the pixels of an image using p rather than A, so we can do a 1-stop increase like this:
convert gradient.png -colorspace RGB -fx "(1-(1-p)(1-p))" result.png
Note that the Wikipedia article, and ImageMagick's -fx both assume your pixel intensities vary between 0 and 1.0. If you are using 8-bit images, you should calculate with 255 in place of 1, namely
+1 stop brighter image = 255-(255-A)(255-A)
or if using 16-bit values
+1 stop brighter image = 65535-(65535-A)(65535-A)
The above fx-based method is however, very slow because the -fx is interpreted rather than compiled, so a faster way to do it is:
convert gradient.png gradient.png -colorspace RGB -compose screen -composite screen.png
Just for fun, another way of looking at that is that we take the inverse of A, that is 1-A, and square it, and then take the inverse, so it can be done like this:
convert gradient.png -colorspace RGB -negate -evaluate pow 2 -negate result.png
The equivalent of -1 stop exposure decrease is to composite the image with itself using the Multiply blend mode, the formula being
1-stop darker image = A x B
which you would do faster with
convert gradient.png gradient.png -colorspace RGB -compose multiply -composite result.png
or even faster, by using memory-to-memory cloning rather than reading from disk twice, with
convert gradient.png -colorspace RGB +clone -compose multiply -composite result.png
but could do equally with
convert gradient.png -colorspace RGB -evaluate pow 2 result.png
Related
Friends,
I have a stack of color-scanned images. Some are from regular white paper with text or images, others were scanned from colored paper (blank pages, same green colored paper used.)
I'd like to identify these colored paper images. Problems:
paper's color ("background") is not scanned very uniformly, often has a wavy or structured pattern
green tone is quite different depending on the scanner used
scanner does not catch the full sheet resulting in a white or shadowed "border" around green area
My idea was to see if say 90% of the image is some sort of green and tried using a sorted histogram. But because of (1) and esp. (2) I have a hard time picking a working color value from the histogram data.
Any help appreciated!
Edit:
Here are three sample images, scanned from the same sheet of paper.
Have a look at HSV colourspace on Wikipedia - specifically this diagram.
It should be a better place to find the colour of your images, regardless of scanner and calibration.
Now, let's create a lime-green, yellow and cyan block and derive its colour using ImageMagick:
magick -size 100x100 xc:lime -colorspace HSV -channel 0 -separate -format "%[fx:mean*360]" info:
120
magick -size 100x100 xc:yellow -colorspace HSV -channel 0 -separate -format "%[fx:mean*360]" info:
60
magick -size 100x100 xc:magenta -colorspace HSV -channel 0 -separate -format "%[fx:mean*360]" info:
300
magick -size 100x100 xc:cyan -colorspace HSV -channel 0 -separate -format "%[fx:mean*360]" info:
180
Hopefully you can see we are correctly calculating the Hue angle. Now to your image. I have added an artificial frame so you can see how to remove the edges:
We can remove the frame like this:
magick YOURSCAN.jpg -gravity center -crop 80% cropped.jpg
So, my complete suggestion would be to crop and convert to HSV and check the mean Hue. You could also test if the image is fairly saturated so it doesn't pick up grey-ish, uncoloured images. You could also test the variance in the Hue channel to see if there are many different colours - or the spread of the hues is large and reject ones where it is large.
magick YOURSCAN.jpg -gravity center -crop 80% -colorspace HSV -channel 0 -separate -format "%[fx:mean*360]" info:
Just for reference, your 3 images come up with the following Hue angles on a scale of 0..360:
79, 68, 73
I would suggest you test a few more samples to establish a reasonable range.
imagemagick's crop command supports cropping to a percentage of an image but the offset values must be specified in pixel values, e.g.:
convert image.png -crop 50%x+10+20
I want to crop with offset values x and y given in percentage of the image width, and height respectively. The pixel values can be calculated, for instance if the image size is 100x200 an offset of 10% would result in 10 and 20 respectively. Is it possible to do this calculation as part of the call to convert? Width and height are available as %w and %h at some places, but this does not work:
convert image.png -crop 50%x+(0.1*%w)+(0.1*%h)
If you're running IM v6 you can use FX expressions with "-set" to set image attributes. By setting the page geometry you can specify the offsets to a calculated percentage and do the crop like this...
convert image.png -set page -%[fx:w*0.1]-%[fx:h*0.1] -crop 50%x+0+0 result.png
That reads the image, sets the geometry for the upper left corner to a location outside the original canvas, and crops to the new top left corner specified by the geometry.
Note the offsets are negative numbers.
Also, if you're doing additional processing in the same command you'll probably want to "+repage" after the crop in order to reset the page geometry to the new WxH+0+0.
Edited to add: You can even include the width and height dimensions for the crop when using "-set page". This command would crop an output of 50% the input width and height, and starting at 10% in from the left and top...
convert image.png \
-set page %[fx:w*0.5]x%[fx:h*0.5]-%[fx:w*0.1]-%[fx:h*0.1] -crop +0+0 result.png
Notice how the crop operation is simply "-crop +0+0" since the dimensions and offsets are in the page geometry.
This method lets you use more complex calculations than just using a percent or number of pixels for the cropped output dimensions.
You cannot do that in ImageMagick 6. But you can do that in ImageMagick 7.
magick image.png -crop "50%x+%[fx:0.1*w]+%[fx:0.1*h]" +repage result.png
In ImageMagick 6, you need to do the computations ahead of the command, store them in a variable and use the variable in the crop command.
However, in ImageMagick 6, you can do the equivalent using -distort with viewport processing as follows:
convert image.png -set option:distort:viewport "%[fx:0.5*w]x%[fx:0.5*h]+%[fx:0.1*w]+%[fx:0.1*h]" -filter point -distort SRT 0 result.png
With v7 ImageMagick, make start image:
magick -size 200x100 gradient: a.jpg
Now crop using lots of calculated widths, heights, offsets:
magick a.jpg -crop "%[fx:w*0.9]x%[fx:h*0.8]+%[fx:w*0.1]+%[fx:h*0.05]" b.png
Check:
identify b.png
b.png PNG 180x80 200x100+20+5 8-bit Gray 256c 408B 0.000u 0:00.000
If you only have v6, use bash and integer arithmetic:
read w h < <(identify -format "%w %h" a.jpg)
convert a.jpg -crop $((w*80/100))x$((h*90/100))+$((w*10/100))+$((h*5/100)) result.png
Check:
identify result.png
result.png PNG 160x90 200x100+20+5 8-bit Gray 256c 412B 0.000u 0:00.000
I want to get the mean of a sequence of images by using Imagemagick. Therefore I use the following command:
convert *.png -evaluate-sequence mean MEAN.png
Each of my images does contain an alpha channel. What I want is: Combine all the images by ignoring the alpha channel.
When I combine the images, the alpha channel is considered in the "mean" method and my final image has transparency. That isn't what I want.
Result:
I tried to add the parameter -alpha off, but then Imagemagick converts the alpha channel to black.
convert *.png -alpha off -evaluate-sequence mean MEAN.png
Result:
Photoshop does it right. I load all images in a stack and create a smart object. When I use the "mean" method in Photoshop, the alpha channel is not considdered in the final result.
Result that I want with Imagemagick:
Does someone have an idea how to do that with Imagemagick?
What you need to do is to use the alpha channels as weighting images for each image. The total fraction of white values at each pixel from all the alpha channels would be the weight to use for the average. So something like this should do what you want or at least be close.
First, cd to your directory of images. Then run the following.
convert *.png -evaluate-sequence mean \
\( -clone 0 -alpha off \) \
\( -clone 0 -alpha extract \) \
-delete 0 +swap -compose divide -composite result.png
This will work if there is some image texture at each pixel coming from al least one image. That is at a given pixel all images are not totally black (transparent).
compare -metric rmse result.png mean_photoshop.png null:
125.167 (0.00190993)
So this shows that there is about 0.2% difference between my result and what you got from photoshop
Maybe this way of working will help you get there - or at least explain the problem:
convert xc:"rgba(255,0,0,1)" xc:"rgba(0,0,0,1)" xc:"rgba(0,0,0,0)" -depth 8 -evaluate-sequence mean txt:
Output
# ImageMagick pixel enumeration: 1,1,65535,srgba
0,0: (21845,0,0,43690) #550000AA srgba(85,0,0,0.666667)
Using IM 6.8.9.4 Q16 or IM 7.0.5.5 Q16 Mac OSX Sierra, this seems to work fine for me:
Make transparent image
convert logo: -transparent white logot.png
Get mean
convert logot.png logot.png logot.png -alpha off -evaluate-sequence mean result.png
magick logot.png logot.png logot.png -alpha off -evaluate-sequence mean result.png
This also seems to work:
convert logot.png logot.png logot.png -channel rgb -evaluate-sequence mean -alpha off result.png
So perhaps you need to upgrade your ImageMagick (and/or libpng?)
Can you post a zip file of some of your input images, so we can test with your images?
One problem that I see is that the PNG images that you provided have black under the transparent areas and not image texture. So when you disable alpha as in my commands above, you see black and the black gets averaged into the final result. Did you use these same PNG images in Photoshop or did you have Photoshop PSD images or some other images that you used and then exported to PNG, which may have put black under the transparent areas. Have you tried using the same PNG images in Photoshop to do the average?
In fact, you have 8-bit color (palette) images, which have one color (black) assigned to be the transparent color.
In ImageMagick convert, I can select a specific color with e.g. -opaque blue. How can I select all grayscale colors (e.g. #000000, #707070, #ffffff)?
Not sure what you are trying to do, but this may help. The greyscale pixels will have a saturation of zero, so that is probably the easiest way to identify them.
First, make a funky sample image:
convert -size 400x100 gradient:black-white -bordercolor red -border 80 image.png
Now make all grey areas (those with very low saturation) transparent:
convert image.png -alpha on -channel A -fx "saturation<0.01?0:1" result.png
Note
Note that the -fx operator is extremely powerful but notoriously slow because it is actually interpolated for each and every pixel. If your images are large, the following technique may be more appropriate.
Basically, I clone the image and convert the whole thing to HSL colorspace and separate the channels. Then I discard the Hue and Lightness channels so I am left with just the Saturation. I then threshold that and copy that back to the original image as the alpha channel. On a 2000x2000 pixel image, this method will run in under a second whereas the -fx method will require 5-6 seconds.
convert image.png \( +clone -colorspace hsl -separate -delete 0,2 -threshold 1% \) -compose copy-opacity -composite result.png
I have image like this from my windstation
I have tried get thoose lines recognized, but lost becuase all filters not recognize lines.
Any ideas what i have use to get it black&white with at least some needed lines?
Typical detection result is something like this:
I need detect edges of digit, which seams not recognized with almost any settings.
This doesn't provide you with a complete guide as to how to solve your image processing question with opencv but it contains some hints and observations that may help you get there. My weapon of choice is ImageMagick, which is installed on most Linux distros and is available for OS X and Windows.
Firstly, I note you have date and time across the top and you haven't cropped correctly at the lower right hand side - these extraneous pixels will affect contrast stretches, so I crop them off.
Secondly, I separate your image in 3 channels - R, G and B and look at them all. The R and B channels are very noisy, so I would probably go with the Green channel. Alternatively, the Lightness channel is pretty reasonable if you go to HSL mode and discard the Hue and Saturation.
convert display.jpg -separate channel.jpg
Red
Green
Blue
Now make a histogram to look at the tonal distribution:
convert display.jpg -crop 500x300+0+80 -colorspace hsl -separate -delete 0,1 -format %c histogram:png:ahistogram.png
Now I can see all your data are down the dark, left-hand end of the histogram, so I do a contrast stretch and a median filter to remove the noise
convert display.jpg -crop 500x300+0+80 -colorspace hsl -separate -delete 0,1 -median 9x9 -normalize -level 0%,40% z.jpg
And a final threshold to get black and white...
convert display.jpg -crop 500x300+0+80 -colorspace hsl -separate -delete 0,1 -median 9x9 -normalize -level 0%,40% -threshold 60% z.jpg
Of course, you can diddle around with the numbers and levels, but there may be a couple of ideas in there that you can develop... in OpenCV or ImageMagick.