Zernike moments descriptor matching based symmetric optical flow for motion estimation and image registration

The conventional optical flow has a fundamental limitation in handling motion details and image registration. In this paper, we propose a Zernike moments descriptor matching based symmetric optical flow estimation for high-quality image registration and motion estimation, which is an integration strategy of descriptor matching of Zernike moments and symmetric optical flow estimation. Zernike moment has less information redundancy and low sensitivity to noise compared to other moments and can well describes the shape characteristics of the objects. Thus, the descriptors obtained by Zernike moments that are defined on the driving points in an image can well reflect the underlying structure. During the computation of descriptors, we hierarchically select the driving points that have distinct attribute features, thus, drastically reducing ambiguity in finding correspondence. Furthermore, a simple and efficient inverse consistency optical flow is proposed with aims of motion estimation and higher registration accuracy, where the flow is naturally symmetric. Experiments implemented on Middlebury beach dataset, MIT dataset and magnetic resonance brain images demonstrate the effectiveness of the proposed method.