基于两尺度分解和特征提取的红外与可见光图像融合

Infrared and Visible Image Fusion Based on Two-Scale Decomposition and Feature Extraction

可见光图像包含大量的纹理信息、梯度信息,而红外图像则包含更多的热辐射信息,通过图像融合可获得更高质量的图像。针对可见光和红外图像融合方法存在目标不突出、图像质量不高的问题,提出一种基于两尺度分解的可见光与红外图像融合方法。首先将源图像进行两尺度分解得到基础层和细节层。接着,基础层采用逐像素比较选取最大值融合规则进行融合得到基础层融合结果。同时,细节层用深度学习网络进行特征提取,分离出不同层次的特征通过L1范数和平均操作,得到稀疏的权重图,再将权重图上采样至相同大小,得到权重图与源图像加权平均得到细节层融合结果。最后,基础层融合结果和细节层融合结果直接相加得到最终融合图像。实验结果表明,我们的方法提出的算法能够较好地提取出待融合特征,在客观指标和视觉效果上都具有良好的表现。

Visible image contains a lot of texture information,gradient information,while infrared image contains more thermal radiation information.Higher quality images can be obtained by image fusion. Currently,many visible and infrared image fusion methods have the problems of unobtrusive targets and low image quality. In this paper,a visible and infrared image fusion method based on two scale decomposition is proposed. Firstly,base layer and detail layer are obtained by two scale decomposition of source image. Next,the fusion results of the base layer are obtained by selecting the maximum fusion rules by comparing each pixel. In the meantime,detail layer adopts deep learning network to extract feature,separate out different levels of features through the L1 norm and average operation,and get the weight map of sparse graph.Then the weight map is transformed to the same size by upsampling. The fusion results of detail layer are obtained by weighted average of the weighted image and the source image. Finally,the fusion results of base layer and detail layer are directly added to obtain the final fusion image. Experimental results show that the proposed algorithm can better extract the features to be fused,and has a good performance in both objective indicators and visual effects.