A question arises as to why create an image search engine based on image content rather than just text. After all, the Google image search engine returns valid results quickly by searching for text only. Also, text-based image search engines are easier to implement and design.
A small (or even large) image search can require searching large amounts of irrelevant web pages, resulting in very little hits, possible irrelevant hits, and wasted time.
It may be hard to describe a specific picture that you are looking for, such as a picture of Galileo around the age of 35 holding a telescope outside around midnight while wearing a white shirt, black pants, and no shoes.
Suppose you are looking for an image of a unique pattern design, such as a specific tie-dye pattern. What would you type in to find it? These types of images can be difficult to find, and specific images can be almost impossible to pinpoint, using a text-based image search.
Again, you are searching for the specific picture of Galileo as described above. There is a very good chance that the level of detail that you want will not be described in the surrounding text of this picture, leaving your search empty when it should not be.