【作者】 周渝人；

【导师】 陈娟；

【作者基本信息】 中国科学院研究生院(长春光学精密机械与物理研究所) ， 机械电子工程， 2014， 博士

【摘要】 随着图像融合技术的发展，红外传感器和可见光传感器的应用领域变得更加的广泛，同时也对红外与可见光图像的融合算法提出了更高的要求。目前，从传感器获得的图像的尺寸越来越大，给像素级融合带来了存储和计算上很大的挑战，压缩感知理论的提出为解决该问题提供了一个方向；传感器在获取图像过程中，难免会受到外接因素的影响，造成获得的图像的对比度变差，可视质量下降，这就要求图像融合算法不仅能够融合互补信息，去除冗余信息，而且还要提高融合图像的对比度，改善视觉效果；融合图像质量评价不但能够评价融合图像的视觉效果，还能评估融合算法性能的优劣，对融合算法的改进和完善，具有指导意义和参考价值。论文对上述三个方面的内容进行了详细研究，主要研究成果如下：1、研究了多尺度分析理论及压缩感知理论，提出一种分立欠采样压缩感知红外与可见光图像融合方法。该方法将小波变换与Contourlet变换结合，获得图像更加稀疏的表达，根据分解后系数分布的特点，设计了分立双放射型采样模式及恰当的融合规则，最后对融合系数进行非线性共轭梯度法压缩感知重构，提高了融合图像的质量。2、针对图像获取过程中可能出现的图像对比度变差、图像可视质量下降的问题，研究了一种对比度增强的红外与可见光图像融合方法。该方法使用双边滤波解决子带分解多尺度Retinex方法在图像增强过程中出现的光晕现象，将改进后的方法应用于红外与可见光图像融合，在完成图像融合的同时，提高了融合图像的对比度，增强了图像的细节。3、针对传统融合图像质量客观评价指标的评估结果与主观视觉感受不符的问题，提出一种符合人眼视觉特性的客观融合图像质量评价指标。该指标在可视信息逼真度的计算中引入对比敏感度作为求和的权重，解决了可视信息逼真度在评价过增强图像时评价值较高的问题。实验结果表明该指标评价结果与主观评价结果一致，而且计算简单，具有实用价值。更多还原

【Abstract】 With the development of image fusion technology, the applications of infraredsensors and visible light sensors become more widespread. Meanwhile, a higherrequirement for fusion algorithms about infrared images and visible images isproposed. Currently, the size of the image obtained from the sensor increases. Withregard to the fusion algorithms for pixel operations, this change takes a lot ofchallenges on the storage and computing. Compressed sensing theory provides a newway to solve this problem. When the sensor is obtaining images, external factorsmay disturber the process. And these factors reduce the image’s contrast. Thus,higher requirements for the fusion algorithm are put forward. Not only can the fusionalgorithm blend the complementary information, but also remove the redundantinformation. At the same time, it is better for the algorithm to improve the visualeffect of the fusion image. The fusion image plays a very important role for thealgorithm evaluation. It can not only make fusion algorithm improved and perfected,but also evaluate the property of the algorithm.The paper has carried lots of studies on these three aspects. There are severalresearch results as follows.1. Through the analysis and research of multi-scale compression perceptiontheory, we proposed a new fusion algorithm on infrared images and visible images.This algorithm is based on the discrete undersampling theory. By combiningcontourlet transform algorithm and wavelet transform algorithm, We can get the image with more sparse expression. According to the characteristics of thecoefficient of energy distribution after transformation, a new discrete doubleradiation sampling mode is proposed. According to the characteristics of thecoefficient of each layer after sampling, the appropriate rules of fusion are used. Thismethod improves the quality of the fused image.2. In the process of image capture, the image contrast may be variation; theimage visual effect may be descended. To solve these problems, a new algorithm isproposed. This algorithm can improve the contrast of the fused image. By using thebilateral filtering, this method works out the halo phenomena appeared in the processof image enhancement of Retinex method with multiscale subband decomposition.This method is applied to image fusion. It not only enhances the image contrast, butalso increases the detail of the image.3. In view of the traditional fusion image, there is a problem for the imagequality. Evaluation results of objective evaluation indicators are different fromsubjective visual perception. In order to solve this problem, a new evaluation indexof fusion image quality is proposed. The index is a kind of objective evaluationcriteria that conforms to the human eye visual features. In the calculation of thevisual information fidelity, contrast sensitivity is introduced as a weight of thesummation. It solves a problem of high value. The value is brought by the visualinformation fidelity when it is evaluating a enhance image. The experimental resultsshow that, the evaluation result is consistent with the subjective evaluation results.And, the algorithm is simple. Therefore, it has a certain practical value.更多还原