基于自适应分解的图像融合及其质量评估问题研究

Image fusion is the key technology to solve the limitation of single imaging sensor and improve the amount of information in images. Now fusion methods based on multi-scale image decomposition have achieved good results in the field of image fusion. However, this kind of algorithm has some defects, for example, artifacts or blurring phenomenon often appears in fused images. The reasons are that most of image decomposition algorithms use fixed decomposition base sets, which cannot guarantee the effective separation of high and low frequency information; fusion rules are simple, and the association among subbands is neglected; the parameter setting in fusion algorithms is lack of the support of theoretical analysis or experimental data. In this project, we will study image fusion using image adaptive decomposition theory, so that the high and low frequency information of source images can be effectively separated. According to the characteristics of high and low frequency subbands and the association among subbands of source images, optimization based image fusion rules are designed. The method of verifying the stability of the image fusion evaluation measures on small sample sets is studied. Combined with the relevance of measures, a construction method of measure set is designed. We will construct a stable and comprehensive evaluation method of image fusion performance to provide the basis for the optimization of image fusion algorithms. Based on the theory above, we further study the typical application of image fusion in medical and remote sensing fields. The implementation of this project will deepen and expand the theoretical research in image processing, information fusion and other fields.

图像融合是解决单一成像传感器局限性、提高图像信息量的关键技术。目前，基于多尺度图像分解的融合方法在图像融合领域内取得较好的效果。但是，该类算法存在一定缺陷，如容易出现伪影、模糊现象。其原因是大部分多尺度图像分解算法均采用某一固定的分解基底，无法保障高低频信息的有效分离；融合规则单一，且忽略子带间的关联关系；算法内参数取值缺乏理论分析或实验数据支撑。针对以上问题，本项目拟研究利用图像自适应分解理论进行图像融合，以便将源图像的高低频信息进行有效分离；针对源图像高低频子带的特点及子带间的关系，设计基于目标优化的跨子带融合规则；研究在小样本上对图像融合评价指标的性能进行验证，结合指标间的相关性，设计指标集构造准则。制定稳定、全面的图像融合效果评价标准，为融合算法优化提供依据。基于以上理论，进一步研究图像融合在医疗、遥感领域内的典型应用。本项目的实施将深化、拓展图像处理和信息融合等方面的理论研究。

图像融合技术可以将多幅图像的显著信息有效地结合起来，并通过一幅图像进行表达，以突破单一成像机制传递信息不充足的局限性。该技术已经广泛应用于医疗、遥感等领域。本项目的研究内容包括图像分解、融合规则和融合质量评估三个问题。为了解决传统图像分解方法造成的高低频成分混叠问题，研究了基于图像本身特性的自适应分解方法及其在图像融合中的应用。解决了自适应分解中的同步性问题。设计了相应的融合规则，融合图像中很好地保留源图像中的显著信息。解决图像融合质量评估指标准确度不高的问题。基于上述理论研究，已发表学术论文23篇(SCI检索论文21篇，EI检索期刊论文2篇)，授权发明专利1项，授权软件著作权3件，培养博士研究生2人，硕士研究生6人，圆满完成了项目的研究目标。