图像融合算法研究及DSP实现

【作者】 吕超峰；

【导师】 程咏梅； 赵永强；

【作者基本信息】 西北工业大学 ， 控制理论与控制工程， 2007， 硕士

【摘要】 多源图像信息融合是自动目标检测与识别领域中一个研究热点，在图像理解、计算机视觉和遥感领域得到了广泛的应用。本文对多源图像信息融合中的关键技术：图像配准及异构图像信息融合算法进行研究，并初步建立了基于DSP的图像融合系统。主要工作如下： 1．对图像信息融合的国内外研究现状进行了综述，给出了图像融合的流程，从算法的特点、适用性和优缺点方面，详细分析了图像配准及像素级图像融合算法，并进行了分类。 2．针对图像互信息配准稳健性和优化时易陷入局部最优的问题，把梯度信息和互信息结合起来作为新的相似性测度函数，利用粒子群算法优化目标函数，在配准中引入了小波多分辨技术，提出了基于粒子群优化的互信息图像配准算法，使得图像配准在提高稳健性的同时，减少了计算时间。仿真结果表明算法是有效的。 3．针对现有的融合算法不能很好地区分噪声和视觉上有意义的特征信息，首先利用二进小波分解后高频系数的局部模极大值得到各尺度的图像边缘，然后利用小波系数的模极大值抑制噪声，结合子带关联和尺度相关的融合准则对去噪后的边缘进行融合，最后基于边缘重构图像，由此，提出了一种基于模极大值和相关性的图像噪声抑制融合算法。算法在抑制噪声的同时更好地保护了边缘特征信息，同时减少了计算量。理论分析和实验结果表明了算法是有效的。 4．针对图像融合中数据量大、传输率高、运算复杂的实际需求，初步设计出了一种基于DM642的实时图像融合方案。软件框架设计采用TI公司最新推出的DSP软件参考框架RF5，采用SCOM和ICC通信机制保证任务间和CELL间正常通信。该系统可支持两路视频采集处理，能进行同构或异构图像的融合。更多还原

【Abstract】 Multi-sensor image fusion is one of the hotspots subject in automatic target detection and recognition, which is widely applied in a variety of filed such as image understanding, computer vision and remote sensing. This dissertation mainly aims at image fusion key technology such as image registration and image fusion, and the image fusion system was primarily built up based on DSP. The main contributions are as follows:A systematic of multi-sensor image fusion theory and methods is reviewed, their feature applicability advantages and drawbacks are indicated generalized. And the development of image fusion’s methods and system are analyzed. The image registration methods are reviewed, and image registration theory and purpose are analyzed.An image registration method based on mutual information is proposed, owning to the fact that the robust of image registration based on mutual information. The gradient flied is combined into mutual information as the new similar measure function. Then the similar measure function is optimized by Particle Swarm Optimization. In order to improve the precision of image registration and reduce the computation time, the wavelet transform is applied in the image registration. Experiment result verifies the effectiveness of the method.An image fusion method based on module maximum and correlation is proposed, owing to the fact that existing image fusion method could not identify meaningful image features from noise. Firstly, after dyadic wavelet decomposition, the image edges of each scale are gotten using the wavelet coefficients’ local module maximum; then, adaptive noise-suppressing method based on the wavelet coefficients’ module maximum is applied to obtain image edges, and the edges are fused, combining the cross-band and the cross-scale correlation; finally, images are reconstructed using the edges. The method not only reduces noise but also preserves edge information, and it can reduce computation as well. Theoretic analysis and experiment result verify the effectiveness of the method.To meet the requirement of image fusion of large quantity data, high transmission ratio and complicated computation, a real-time image processing system更多还原