White Paper WHP 177 Download
This White Paper describes the operation of CLARET, an image CLAssification and RETrieval system. One method of identifying image content is implemented and used in two different scenarios. The first scenario is object classification from five learned classes (pedestrians, cars, motorbikes, bicycles and rocket propelled grenades). Images are analysed in terms of the learned classes resulting in a confidence factor that the object class is present in the image. This scenario can be used to automatically generate keyword metadata from images. The second scenario is image retrieval (search) which visually orders images according to their similarity to a selected image. This is known as 'query by example'. Query by example can be used to search through large image archives or to prioritise imcoming UGC (User Generated Content). CLARET is a collaboration project between the Unveristy Surrey and BBC R&D.