Image Classification - Detecting an image is cartoon-like

Question

I have a large amount of jpeg thumbnail images ranging in size from 120x90 to 320x240 and I would like to classify them as either Real Life-like or Cartoon-like.

Are there any applications that will have cartoon classification capabilities?

This application should work on Linux, and should take an image path on the command-line and return either 0 or 1 (echo $?).

Answer 1

I think you could do something like this with Imagemagick. It has image quantization and histogram analysis features that you'll probably need to give this a real treatment.

The simplest thing to do is count the number of unique colors in each picture - cartoons should generally have fewer than photos. This may work as is if your search space is fairly simple. i.e. differentiating simple cartoons form color photos. If you have 'fancy' cartoons, you may have to add additional checks. I added an extra echo for RGB vs. Grey color space before checking each image.

A more sophisticated test might involve checking the histogram, either total or in RGB space of each image.

#!/bin/bash

for i in `ls *.jpg`
do
    echo "$i is `convert $i -format \"%[colorspace]\" info:`"
    x=`convert $i -unique-colors txt:- | wc -l`

    if [ $x -le 512 ]; then
        echo "$i is cartoon like ($x)"
    elif [ $x -le 1024 ]; then
        echo "$i is a bw photo ($x)"
    else
        echo "$i is real life-like ($x)"
    fi
done

The main complication is separating complex computer animations from b&w photos. A B&W photo may have relatively few unique colors in it, while a sophisticated cartoom may have thousands due to computer aided shading. You'll probably need to experiment with the thresholds for 'X' depending on what your images look like.

Answer 2

Provided that you have images that you can show to school children and there are lots of cartoons.

You can go to a school and use HUMAN COMPUTING.