PROJECT TITLE :

Randomized Spatial Context for Object Search

ABSTRACT:

Searching visual objects in massive image or video information sets is a difficult drawback, because it needs efficient matching and correct localization of query objects that always occupy a small half of a picture. Though spatial context has been shown to assist turn out additional reliable detection than methods that match native options individually, how to extract acceptable spatial context remains an open downside. Rather than using mounted-scale spatial context, we propose a randomized approach to deriving spatial context, in the shape of spatial random partition. The effect of spatial context is achieved by averaging the matching scores over multiple random patches. Our approach offers three edges: 1) the aggregation of the matching scores over multiple random patches provides sturdy local matching; two) the matched objects can be directly identified on the pixelwise confidence map, which results in economical object localization; and three) our algorithm lends itself to easy parallelization and additionally allows a versatile tradeoff between accuracy and speed through adjusting the number of partition times. Each theoretical studies and experimental comparisons with the state-of-the-art strategies validate the benefits of our approach.