一种改进的在线图像对齐算法

An improved online image alignment algorithm

摘要:

由于图像之间存在光照变化巨大、部分遮挡和损坏等现象，稳健且高效的图像对齐仍然是一项具有挑战性的任务。为此，提出了一种改进的在线图像对齐算法。通过图像梯度方向(IGO)的主成分分析(PCA)来提供比像素强度更可靠的低维子空间，然后在IGO域中寻找对齐，使新到达图像的IGO对齐被分解为稀疏误差和从先前良好对齐图像上学习得到的IGO-PCA基的线性组合的总和，进而图像对齐问题可被建模为l1范数最小化问题。将该问题松弛为凸优化问题，并提出一种基于乘子交替方向法的凸优化求解算法。考虑IGO均值的偏移，基于增量奇异值分解自适应地更新IGO-PCA基。在大量具有挑战性的数据集上验证了所提算法的有效性。实验结果表明，相比于目前典型的尺度不变特征变换(SIFT)算法、稀疏和低秩分解(RASL)算法和变换Grassmannian鲁棒自适应子空间跟踪算法 (t-GRASTA)，本文算法的对齐效果更佳，对于图像的光照变化和遮挡现象等具有更强的鲁棒性。

Abstract:

Due to the phenomena of huge illumination changes, partial occlusion and damage between images, robust and efficient image alignment is still a challenging task. An improved online image alignment algorithm is proposed. Firstly, the Principal Component Analysis(PCA) of image gradient orientations(IGO) is utilized to provide a lower dimensional subspace which is more reliable than the intensity of pixels. The alignment is sought in the IGO domain such that the aligned IGO of the newly arrived image can be decomposed as the sum of a sparse error and a linear composition of the IGO-PCA basis learned from previously well-aligned ones. The image alignment problem can be modeled as a norm minimization problem. The problem is relaxed to a convex optimization problem in this paper, and a convex optimization algorithm based on multiplier alternating direction method is proposed. IGO-PCA basis are adaptively updated based on incremental singular value decomposition considering the migration of IGO mean in this paper. The effectiveness of the proposed algorithm is validated on a large number of challenging data sets. The experimental results show that compared with the current typical SIFT algorithm, Robust Alignment by Sparse and Low-rank decomposition(RASL) algorithm and transformed Grassmannian Robust Adaptive Subspace Tracking Algorithm(t-GRASTA), the alignment effect of the proposed algorithm is the best, and it has the strongest robustness to illumination changes and occlusion phenomena of images.