It is possible to do so by using a mapping node, the thing is that you may lack of precision since this job is done usually proceduraly without messing with normalized values (as UVs) but absolute instead (pixels).
So the idea is to create a plane (or as many as the poses you want to "extract") and with a mapping node pick and fit the proper sprites on them using the spread sheet as an image texture.
A more elegant and efficient way would be to do this automatically by using a python script but then it becomes a pain in the a#s since there is no any bundled / native lib for image processing.
In theory what you had to do would be to parse the JSON file (as is in your example) and get the directions from there for the pixel width/height and the position on the spread sheet of each pose or element.
you may take a closer look on this post : Crop image with Python script
where is the basic concept of cropping images in blender with python scripting.