With advances in digital imaging, the amount of digital images will increas
e tremendously. To locate relevant images in a large collection of images p
resents a challenging and genuine problem fur content-based retrieval resea
rch. This paper presents a novel framework called visual keywords for image
indexation and query formulation. Visual keywords are flexible and intuiti
ve visual prototypes specified perceptually from sample domain images. A vi
sual content is described and indexed by flexible spatial aggregation of th
e soft presence of visual keywords. A new query method based on visual cons
traints is also proposed to allow direct and explicit. content specificatio
n. Last but nor least, we have developed a digital album prototype to demon
strate query and retrieval on both home photos and stuck photos based on vi
sual keywords.