Identify Different Types of Images

[Vanns]

I am having images in different formats like screenshots, images having handwritten/lines/shapes and pictures (nature/person/animal). How to identify them individually. Is there any way to do that?

[fmw42]

This post was not relevant to the topic. So I have moved it and changed the title.

[snibgo]

Is there any way to do that?

There are various techniques for identifying whether a photo contain people or animals etc. These are not simple techniques, and often involve "training" a system.

For identifying whether the image is a screenshot, or hand-written shape, or photo, the process may be simpler. Do you have example images?

[Vanns]

Picture is


Screenshot is


Line drawing is

[snibgo]

Code: Select all

%IM%convert screenshot-gimp-windowsvista.png -statistic standard-deviation 3x3 -grayscale Brightness -fill White +opaque Black -negate -format %[fx:mean] info:
0.337484

So about one-third of the screenshot is flat colour. For the line drawing, about 63%. For the photo, only 1%.

(Photos have virtually no areas of flat colour.)

The third image is grayscale with a very high mean (96%).

[Vanns]

For what parameter this value is evaluated? Can u please explain the command?

One third of the screenshot mean its size or?

[snibgo]

%IM%convert screenshot-gimp-windowsvista.png : read the image.

-statistic standard-deviation 3x3 : for every pixel, for each colour channel, calculate the standard deviation of a 3x3 square centred on the pixel. Where we have a flat colour, all pixels are the same, so SD=0.

-grayscale Brightness : find the largest of all the colour channels.

-fill White +opaque Black : where a pixel isn't black, make it white. So any pixels with any SD > 0 will become white.

-negate : makes black into white and white into black. Now, any pixels with any SD > 0 will be black. Pixels with all SD=0 are white.

-format %[fx:mean] info: : Find the mean. As all pixels are either black or white, this is the proportion of pixels that are white.

So the answer, 0.337484, means that one-third of the pixels are the centre of 3x3 pixel squares where all 9 pixels are the same colour.

[Vanns]

But when I tried with my laptop's screenshot, it is showing 0.886308

It is not around 0.33. What could we do in these cases?

[snibgo]

Photos will have virtually no areas of flat colour, around 0.01. Screenshots and line drawing will have much higher numbers, perhaps between 0.1 and 1.0.

[Vanns]

How to make difference between screenshot and line drawings depending upon this value?

[snibgo]

I don't understand the question.

[Vanns]

Can u help me with jmagick function calls for the same command?

[snibgo]

Sorry, I don't use jmagick.

[karlw]

But when I tried with my laptop's screenshot, it is showing 0.886308

It is not around 0.33. What could we do in these cases?

you not screen your wallpaper or open graphic program in case?
coz your example for screenshot is actualy 50% "Picture" and 50% drawed windows.
i'm affraid screenshot what guis tok about is open Exel or World.

[sandy]

convert screenshot-gimp-windowsvista.png -statistic standard-deviation 3x3 -grayscale Brightness -fill White +opaque Black -negate -format %[fx:mean] info:
0.337484

Can we find out the same value using jmagick?