A Novel Mismatching Elimination Algorithm Based on Catadioptric Panoramic Images

Catadioptric panoramic images are gaining increasingly attention in the field of computer vision. However, these images’ characteristics of complex projection imaging and distortion make most mismatiching elimination algorithms difficult to be applied to the ring shaped catadiptric panoramic images directly. In order to solve the above problems, a mismatching elimination algorithm based on azimuthal angles of features is proposed in this paper. The novel algorithm is inspired by the warping visual homing algorithm. It firstly makes use of azimuthal angles of features in one image to estimate the azimuthal angles of their corresponding ones in its matching image. Then it eliminates the matching feature pairs whose azimuthal angle deviation is too large by comparing the estimated value and the actual value. The novel algorithm is proved to be effective by extensive experiments based on panoramic image databases.