Being able to do optical character recognition to filter images would help, since most images aren't described or may not provide a good description of the text in an image.