What is MulticolorEngine?

MulticolorEngine is an image search engine that finds images based on their common colors.

Large collections of images have traditionally been organized by keyword. Images are examined and annotated with descriptive keywords, and the keywords assigned to each image are stored in a keyword index (along with a reference to the image). This index is then used to support keyword queries.

A major problem for such image management systems is that there is no universally accepted way of classifying most types of images. For instance, plant images will be classified very differently by a landscaper, a doctor studying medicinal herbs, a nutritionist and a plant pathologist. Thus, in many cases, it is difficult to construct a keyword index that allows an arbitrary user intuitive access to images of interest.

MulticolorEngine is not intended as a replacement for a traditional keyword search engine. Instead, it is meant to complement an existing keyword search engine, resulting in a system that offers greater flexibility in identifying images of interest.

A typical use case would be to supply keywords to obtain a set of images matching some criteria, identify a specific image of interest and then use multicolor search to query the system for images containing the same colors. Another use case is to perform a combined color and keyword search by uploading an image along with a set of keywords. And the last use case is to peform a pure color search using an input image or a set of input colors.

MulticolorEngine has the potential to save you a great deal of time by querying the system for images of interest rather than manually searching for relevant images by browsing through large image sets.

How does it work?

The MulticolorEngine API must create an index before it can search an image set. Each image in the index has a digital fingerprint attached to it, which is stored in an internal database. The fingerprint contains special information about the visual features of each image, such as colors and their distribution across the image. The index allows images to be compared quickly by eliminating the need to re-examine the images in the set. The MulticolorEngine API can then efficiently compute a relevance score which measures how similar the color distribution is across two images.