19

My friend has an colored image with Chinese handwriting (basically by taking photo of or scanning what he wrote on a piece of white paper), and he would like me to convert it into a black and white binary image. Are there applications under Ubuntu that can accomplish that?

Here is an example image:

enter image description here

rahmu
  • 20,023
Tim
  • 101,790

5 Answers5

32

What you want is referred to as "threshold" in image processing. Basically, it takes an image as an input and outputs an image that has all pixels with a value below a given threshold set to black, and all pixels the value of which is above the threshold set to white. This results in a black-and-white image from an arbitrary input image.

Generally, you want to convert to grayscale first for more predictable results, but it is possible to threshold a full-color image as well.

You can use a graphical tool such as GIMP to do this interactively (you'll find the tool through the main menu -> Colors -> Threshold), or you can use ImageMagick something like this:

convert colored.png -threshold 75% thres_colored.png

Running the above command on the example image produces the result shown below.

Black-and-white version of OP's image

Since thresholding is often somewhat of a trial-and-error process to get a result you're happy with, particularly if the source image is not very close to black-and-white already, I recommend the GUI approach if possible, but if that is not an option for whatever reason you can do it through the command line as well. For finer control of the output, you can use tools like color curves, levels and contrast first to isolate the light and dark portions of the image better before thresholding. (Actually, threshold can be seen as an extreme case of using the color curves tool.)

user
  • 28,901
  • 2
    Thanks! Since a colored image has RBG three channels, what does/can the threshold apply to in general? – Tim Jan 09 '14 at 23:00
  • @Tim It probably depends on the software, but I would expect the threshold (unless you specify per channel or for a specific channel, see e.g. ImageMagick's convert's -channel option) to be applied to some sort of "value" of the pixel, which is computed from all channels. That's the reason why I said you might want to convert to grayscale first for more predictable results. (Also see my edit.) – user Jan 10 '14 at 08:17
  • Thanks! Does there exist some documentation for what the threshold apply to in the command shown in your post? – Tim Jan 10 '14 at 18:29
  • @Tim Not really. I expected convert to take either a percentage of the maximum value (which should have been 256 per channel) or a specific value, but I could only get a useful result when specifying a percentage. When you do it with a graphical tool, including GIMP, you'll generally have a histogram that shows the tonal distribution of the image; that will be a great aid in picking the proper value. Using only the command line, unless you have a specific reason to do so, is probably more trouble than it's worth, really. – user Jan 10 '14 at 18:31
  • 2
    As a side note, there exist other thresholding methods that don't have to rely on a hard-coded threshold level. For example, ImageMagick includes -lat which performs a local adaptive threshold, taking into account the surrounding pixels. – voithos Jan 15 '14 at 22:37
7

You can use Imagemagick: 

convert test.png -colorspace Gray gray_colorspace.png

From here.

Here is what I got after applying to your image:

enter image description here

Ketan
  • 9,226
  • 12
    "Binary" comes from "bi" meaning "two", so I assume the OP wants to convert the image to pure black-and-white. Converting to grayscale yields a lot more than two levels. – user Jan 09 '14 at 20:57
  • Agreed! makes sense, +1. – Ketan Jan 09 '14 at 21:31
7

ImageMagick convert -monochrome

The -monochrome option uses some smart dithering, and makes the output much more visible than -threshold if you intend it for human consumption:

convert -monochrome signature.png out.png

enter image description here

Does not make much difference for such a simple image, but for larger ones, it is striking:

Ciro Santilli OurBigBook.com
  • 18,092
  • 4
  • 117
  • 102
2

You can also do this easily with the netpbm toolkit:

anytopnm inputfile | ppmtopgm | pgmtopbm > outputfile

ppmtopgm converts to a grayscale image, pgmtopbm converts to a black-or-white image, and we then redirect the output to a file. It will be in the pbm format; if you want something more common, you'll have to add an output convertor (e.g., pnmtopng or some such)

Wouter Verhelst
  • 9,331
  • anytopnm inputfile | ppmtopgm | pgmtopbm -thershold > outputfile works so much faster than convert -threshold 50% inputfile outputfile! (And avoids the latter’s resource constraints and “security” policies.) – Amir Feb 09 '21 at 06:49
1

Otsu's method will perform better when dealing with texts over a background with slightly gradient colors:

https://stackoverflow.com/questions/65945662/how-do-i-convert-a-color-image-to-black-and-white-using-imagemagick/65999445#65999445

According to @fmw42 's comparison on multiple threshold methods: the Local Adaptive(its equivalent in imagemagick is -lat) algorithm might work best on texts over background:

Also try to combine with connected components processing that was introduced in another answer made by @fmw42 to remove unnecessary edges/dots which might get confused by any further OCR process.

n0099
  • 113