结合归一化差分高斯特征的图像匹配技术研究

Image matching technology research based on normalized DOG features

图像匹配技术是目前国际上航空摄影测量、光学和雷达跟踪、飞行器巡航制导、导弹投射系统的末制导寻的等领域发展迅速前沿、尖端技术之一。近年来,SIFT算法以其优秀的匹配性能在图像匹配领域受到广泛的关注。鉴于传统SIFT及改进算法计算复杂,难于实际应用的问题,提出结合归一化差分高斯特征的图像自动匹配技术,提取对于复杂影响鲁棒的归一化差分高斯特征,通过在特征点处建立主方向趋势特征向量,减少传统特征向量的冗余,解决不同条件下获取的目标或场景的图像匹配问题.大量实验表明,本匹配算法能够适应图像的旋转、缩放、平移等变化,并且对于噪声、光照变化影响鲁棒;同时计算速度快,具有较好的工程实用价值。

Image matching is the key problem in the airborne optical measurement,optical and radar tracking,terminal-guiding areas.In these years,SIFT algorithm gets more attention due to its excellent matching performance.However,traditional SIFT and some improve ones are hard to apply in the practice because of their high complexity.The paper proposes a new method that uses normalized DOG features to enhance the matching and construct key orientation trend vector to reduce the traditional redundancy.Now,the algorithm is applied in the hardware system.Lots of experiments show that the method can detect the target exactly to overcome the variations of the scale,rotation,translation,blur and the small distinction between the images due to the different condition.The algorithm time is less than 200 ms,which can satisfy the practical need by using two DSP parallel computing.