I have no experience with it but I think the main problem is to define what belongs to the object under consideration. For example if you want to compute the fractal dimension of a werewolf you have to define which pixel is part of the werewolf and which isn't. Simple methods might be that you click on one point on the image and the algorithm counts all pixels with the same color using floodfill, or overall, all pixels of the same color. Another prerequisite might be that you convert the image to black-white, with black defining parts of the object and white not.
You can follow below steps in analyzing the Fractal Dimension of an image using box counting method in ImageJ software.
1. Set the scale of values for a known unit (cm, mm etc.)
2. Convert the image into 8 bit
3. Threshold the images
4. Then follow the path of,
Analyze, Tools, Fractal Box Count
5. Insert the required box sizes
Then you get the slope of the least-squares linear regression of the log (box count) vs. log (box size) plot along with the Fractal Dimension of the image.