结合仿射变换和多层B样条配准的湍流畸变图像校正

Turbulence distorted image correction using affine transformation and multilevel B-spline registration

为解决大气湍流引起的图像抖动和图像偏移问题,提出了一种结合仿射变换和多层B样条配准的图像校正方法。由于湍流像素偏移具有非线性和随机性,且序列中常伴有成像系统运动引起目标全局运动,该方法将像素校正分为整体运动校正和局部非刚性配准两部分。采用仿射变换描述目标整体运动;引入多分辨率策略处理像素偏移,利用多层B样条对湍流引起的局部形变进行非刚性配准。基于对称约束的代价函数,通过梯度判定感兴趣区域,进一步细化控制点网格提高B样条配准精度。最后,采用有限内存拟牛顿法优化代价函数,得到像素偏移量,实现畸变图像的像素校正。对真实图像和实际序列图像分别进行了实验,结果表明:该方法能够有效降低湍流造成像素偏移畸变,在噪声和模糊干扰下校正结果依然理想。

A correction method for distortion sequences was proposed based on affine transform and multilevel B-spline registration to overcome the image distortion and pixel deviation from the atmospheric turbulence.As the pixel deviation caused by turbulence is nonlinear and random,and the global motion caused by the imaging system movement might exist in practice,the correction method divided the pixel into two parts,the global motion estimation and the local non-rigid registration.The affine transformation was adopted to eliminate the global motion.The multi-resolution strategy was introduced to process pixel distortion and the multilevel B-spline was adopted to register the local motion.On the basis of the symmetry constraint cost function,the control point gridding was fined further to improve the B-spline accuracy by evaluating the regions of interest with the gradient method.Finally,limited-memory Broyden-Fletcher-Glodfarb-Shanno（L-BFGS）algorithm was used tominimize the cost function to obtain the pixel deviation and to complement the correction of image distortion.Experiments demonstrate that the proposed method effectively reduces the pixel distortion from the atmospheric turbulence even for noisy and blurred conditions.