There are now also a growing number of systems which use computer vision technology for the automated identification of biological objects (individuals) and/or groups (e.g. species, guilds). Typically these systems are used by non-taxonomists (e.g. ecologists, pest control officers, parataxonomists) to rapidly identify specious tropical biota (e.g. parasitic wasp in Costa Rica). In Algorithmic terms, these systems fall into two broad classes:


Holistic systems which use the entirety of the presented image, or of some region of interest thereof to make an identification. These systems are typically based on principal component analysis, self organising maps (e.g. the plastic self organising map, PSOM), nearest neighbour correlation (NNC) or some form of artificial neural net (ANN). Examples of systems of this kind include DAISY which uses a hybrid PSOM/NNC approach and SPIDA which uses a modified variant of the back propagation neural network.


Feature based systems: These sorts of system extract features from the input imagery and then use these features for subsequent recognition. Examples of this kind of algorithm include the ABIS (Automated Bee Identification System) from Bonn University and the WEKA system from the University of Waikato in New Zealand. Although this type of system may achieve accuracies which are marginally superior to those achieved by the holistic systems, they are intrinsically less flexible. For example ABIS is restricted to identifying insects such as bees and flies which have membranous wings, and in the case of earlier versions of the system at least, significant operator expertise was needed. Both of the holistic approaches cited are essentially generic. For example, in addition to insects, the DAISY system has also been used to classify human faces, foraminifera, bones, aircraft contrails and (with suitable pre-processing) even sounds, all with some measure of success.


The images above show the DAISY system in operation, identifying a specimen of the Belizian Sphingid Cocytius duponchel (Poey, 1832) which was caught in a light trap at the Las Cuevas Field Station in Belize. In order to normalise input imagery for the effects of scale and pose (specimens are live imaged using a digital camera) a PolyROI (polygonal region of interest) is drawn around the part which is being used to identify the specimen, in this case the wing (upper image). Once DAISY has identified the specimen it spawns an HTML browser which points to a URL providing information on the specimen it has identified (lower image). The DAISY backend is flexible: if there is no canned URL available, the backend system can automatically query web search engines such as google for appropriate information,