Endoscopic optical coherence tomography (OCT), and other imaging modalities that use a mechanically rotated probe, often suffer from image degradation due to non-uniform rotation distortion (NURD). In this paper we present a new method to align a sequence of images by globally optimizing the match between individual lines in subsequent frames. It uses dynamic programming to find a continuous path through a cost matrix that measures the similarity between regions of two frames being aligned. The path represents the angular mismatch corresponding to the NURD. The prime advantage of this novel approach compared to earlier work is the line-to-line continuity, which accurately captures slow intra-frame variations in rotational velocity of the probe. The algorithm is optimized using data from a clinically available intravascular OCT instrument in a realistic vessel phantom. Sensitivity of the performance to imaging and optimization parameters is investigated using a computational phantom. Finally, the algorithm's efficacy is demonstrated on an in vivo recording inside a human coronary artery, exhibiting strong motion artifact.