Research of Motion Estimation Algorithm Based on Digital Images
|Keywords||image processing motion estimation image registration phase correlation image restoration real-time stabilization|
With the development of imaging technique and computer technique, Digital image processing technique is used in more and more areas, such as bio-medicine, information science, astronomy, environmental resource science. Those new-born applications usually have higher requirements to image processing algorithms. As the motion estimation algorithm is one of the fundamental techniques to the image processing techniques, which contain image stitching, image restoration, real-time image stabilization, super-resolution reconstruction and high dynamic contrast ratio building, etc. Motion estimation is playing an important role in modern image processing. This paper is mainly about motion estimation algorithms.Firstly, the background and significance of this research are introduced. Then we discuss about the model of image motion, analyze and implement several widely-used motion estimation algorithms that based on image pixel information and image feature. This paper also introduces several methods used to improving algorithm performance, including sub-pixel acquirement, algorithm efficiency optimization and robust improving. These methods are applied to specific motion estimation algorithms and significantly improve performance of the algorithms. Based on existing methods and our analysis, this paper presents a weighted phase correlation algorithm for motion estimation. The new method suppresses the effect of random noise and non-ideal sampling by using windows and low-pass filters. Besides, in order to improving the SNR (signal to noise ratio) of correlation peak, a weighted strategy is added to frequency spectrum. At the algorithm testing part, we designed an experiment which contains simulated and real-taken images to test and compare the performances of different algorithms. The experiment can objectively and comprehensively reflects the characteristics of algorithms. At the end of this paper, motion estimation algorithms are applied in two laboratory systems, which are motion-blurred image restoration system and real-time stabilization system.This paper has three main innovations. First, several methods which can improve algorithm performances are proposed, especially for the method of suppression of the random-noise and non-ideal sampling effect. Second, a new motion estimation algorithm is presented, which based on classical phase correlation algorithm. Experiments prove that this method is excellent. Third, we provide a novel method to calculate blur kernel from motion estimation results.