This paper mainly focuses on how to eectively and eciently measure visual similarity for local feature<br> based representation. Among existing methods, metrics based on Bag of Visual Word (BoV) techniques<br> are ecient and conceptually simple, at the expense of eectiveness. By contrast, kernel based metrics are<br> more eective, but at the cost of greater computational complexity and increased storage requirements.<br> We show that a unied visual matching
