Home  |  Project  |  E-mail  |  Links
	Project
				Content based image retrieval (CBIR) is an exciting and indepth area of research which has garnered much interest over the past several years. It is a procedure in which a set of images is retrieved from a given image database, which are perceived to be similar to a specific query based on visual content. With the introduction of digital photography and the growth of the internet, combined with today’s computers capable of storing and processing multimedia content much more quickly and efficiently the need for such retrieval systems is obvious. This necessity has generated in vast amounts of research into the area, resulting in a wide array of CBIR systems currently available for commercial and research purposes. Approaches towards CBIR however, generally abide to a similar methodology, combining aspects of image processing, pattern recognition and supervised learning.  As CBIR systems evolve it is becoming increasingly obvious that in order to provide efficient retrieval it is necessary to include the human in the loop. There is still a discrepancy between perceptual and computation results of image retrieval due to inconsistencies between what the user and the system regard as the relevant features of the query. An issue is how to derive a weighting scheme in order to balance the relative importance of different feature types. At present, there are two key methods in approaching this issue. One technique involves applying weights to the most predominant features during the calculation of feature distances based on the contents of the image database. However, a far more interesting method which could be employed is to essentially incorporate the user themselves in the retrieval process. Such weighting methods utilizing user interaction are derived on the concept that different users may have different perceptions of similarity and thus allowing the system to continually develop a set of criteria which matches the user’s perception of similarity.
Abstract
Sample Results
Log
Test Data & Source Code
Documents