时间图像序列配准的研究

STUDY ON REGlSTRAT10N OF TEMPORAL IMAGE SEQUENCES

本文分析了时间图像序列配准的特殊性，提出了应从以下五个方面来设计时序图像的配准：预处理以补偿外部条件的变化、分析时序图像的不变量以确定代价函数的项、将目标物分解为几部分从而最终确定代价函数、选择较优的算法以提高配准速度。研究了甲襞毛细血管的时序图像配准，分析了甲襞毛细血管时序图像的不变量，提出了多模板－序列阈值配准算法，实验表明，此算法能成功地完成甲襞毛细血管的时序图像配准。

Analysis of characteristics of registration of temporal image sequences showed that 5 main aspects should be considered to design a fast registration algorithm with high accuracy: 1)preprocessing to compensate variations of external conditions, 2) analysis and 3) determination of invariances in temporal image sequences to ascertain items of penalty function, 4) separation of object to be registered into several parts for full determination of penalty function and, 5) selection of optimal algorithm to perform fast registration. Using our theory of design, registration of nailfold capillary temporal image sequences was studied. After analysis and selectipn of invariences of nailfold capillaries, a multi-mask registration algorithm was suggested with adaptive threshold series, and results showed that the new algorithm performed fast and precise registration of nailfold capillary temporal images.