Translation-based digital image stabilization techniques

Abstract

Image stabilization is a modern technique in video industry, and it can be divided into optical, electronic, and digital image stabilizations (OIS, EIS and DIS). Among these, DIS contains more advantages than the others as it can apply on/off-line applications, minimize hardware implementation, and combine with other systems to enhance the fidelity of stabilization. The basic elements of the DIS system are the motion estimation (ME) and motion correction / motion compensation (MC). Since undesired global motion is mostly translationbased, many stabilization techniques aim to tackle with this motion only. Such systems are also referred as two-dimensional (2D) stabilization techniques. In this project, some of the translation-based DIS techniques are studied and implemented in post-processing purpose by the use of MATLAB 7. For the evaluation, peak signal-to-noise ratio (PSNR) a key factor that can reflect the superposition of the two adjacent images. The highest PSNR, the most perfect image stabilization will be.