Advanced digital image stabilization using similarity-constrained optimization

As many people have portable video devices such as cameras on cell phones and camcorders, image stabilization technique is a crucial and challenging task in computer vision applications, and many image stabilization techniques have been researched over many years. We propose a digital image stabilization method that only uses a software algorithm without additional hardware devices. Furthermore, a novel digital image stabilization method composed of three steps that use similarity-constrained nonlinear optimizer is introduced and applied to many unstabilized videos. First, a feature detection technique called moment-based speeded-up robust features (MSURF) is utilized to obtain the transformation matrix. Second, the k-means clustering algorithm is used to detect and remove some of the outliers that cause residual errors during feature matching. Third, the transformation matrix is optimized using nonlinear optimization algorithms to maintain the similarity of the transformation matrix. The experimental results prove that the proposed algorithm provides accurate image stabilization performance.