Traditional Search Engines work on matching your search terms to tags attached to a content. Such tags could be automatically populated or manually entered against the product or any other content that is searched on. The problem is that your search relevancy is a function of correct tagging of the content. For e.g: If i have a striped Shirt and i search for a striped Shirt, i will not get it unless somebody tags the shirt with the attribute that it is striped in type.
Image Search works differently and it does not required any such tagging of content before the content is searchable. It compares searchable image with the actual image and show the matched result.
We developed such a visual search engine for an Apparel retailer based out of UK and were able to successfully demonstrate an improvement of search results.