Dual tree fractional quaternion wavelet transform for disparity estimation.

摘要: This paper proposes a novel phase based approach for computing disparity as the optical flow from given pair of consecutive images. A new dual tree fractional quaternion wavelet transform (FrQWT) is proposed by defining 2D Fourier spectrum upto single quadrant. In FrQWT, each consists real part (a DWT wavelet) and three imaginary parts that are organized according to algebra. First two FrQWT phases encode shifts image features in absolute horizontal vertical coordinate system, while third has texture information. The allowed multi-scale framework calculating adjusting local disparities executing unwrapping coarse fine scales with linear computational efficiency.