Study On Key Algorithms For Image Sequence Processing

Infrared imaging technology has expanded the scope of human cognition and extended human vision to invisible infrared spectrum which plays a more and more important role in modern military and civilian fields. However, infrared imaging system is limited by its mechanism and device characters which lead to low resolution and low contrast. In the meantime, since infrared imaging systems are widely used on moving platform, the complex environment greatly affects image quality as well. Image processing technique for infrared image sequences is always an important aspect in infrared imaging research area.This paper studied infrared image processing technologies and proposed several sequential infrared images processing algorithm focusing on image enhancement, image stabilization and colorization of infrared image sequence:As a classic human vision theory based color equilibrium theory, Retinex theory works well on color image enhancement but not many achievements in infrared image processing field. In this paper, a modified multi-scale adoptive Retinex algorithm is proposed. In this algorithm, the surround space constant automatically changes based on image details.In many cases, image enhancement algorithms are required not only to improve contrast but also to adjust overall dynamic range. In this paper, the ’halo’ phenomenon is carefully studied and introduced bilateral filter (BF). An infrared image enhancement algorithm based on Retinex theory and BF is proposed. This algorithm could enhance contrast, compress dynamic range and avoid ’halo’ phenomenon at the sametime.Based on working properties of imaging devices on moving platform, this paper studied causes and properties of jitters in vedio sequences. Geometry model of infrared imaging and motion model of image sequences are discussed and focusing on motion estimation and motion filtering. We studied several common used registration algorithms in motion estimation and introduced BRISK algorithm, then a noval BRISK based feature point matching algorithm is proposed to achieve better global motion estimation. Furthermore, several motion filtering algorithms are discussed and Kalmanpartical filter based motion filtering algorithm is proposed.Based on analysises of facts that affect color transfer algorithms, several modifications for traditional aogorithms are made to achive better colorization for infrared images. Since the choice of color space and transfer algorithm greatly affect the outcome of colorization algorithms, we introduced LLE algorithm to propose a novel color transfer method based on LLE algorithm and verified by experiments.