Improved binary feature matching through fusion of hamming distance and fragile bit weight

In this paper, we define a new metric, the Fragile Bit Weight (FBW), which is used in binary feature matching and measures how two features differ. High FBWs are associated with genuine matches between two binary features and low FBWs are associated with impostor ones. One bit in binary feature is deemed fragile if its sign of value reverses easily across the local image patch that has changed slightly. Previous binary feature extract algorithms ignore the fact that the signs of fragile bits are not stable through image transform. Rather than ignore fragile bits completely, we consider what beneficial information can be obtained from those fragile bits. In our approach, we exploit FBW as a measure in binary feature match to remove the false matches. In experiments, using FBW can effectively remove the false matches and highly improve the accuracy of feature match. Then, we find that fusion of FBW and Hamming distance work better in feature matching than Hamming distance alone. Furthermore, FBW can easily integrate in the well-established binary feature schemes if those descriptor bit in extract from comparison of pixels.