多分辨率分析耦合近似稀疏表示的图像融合算法

为了更好地处理图像高维特征奇异性,并兼顾融合图像目标特征与平均强度信息,提出了一种多分辨率分析与近似稀疏表示的图像融合算法。首先,对源图像进行对尺度分析,分别得到图像的高频和低频信息;然后,设计了近似稀疏表示(sparse representation,SR),通过近似SR系数来表示图像高频信息和低频信息;并利用绝对最大选择技术对近似SR稀疏转换,得到低频子带的近似系数和高频子带的细节系数,以达到用最少的系数来逼近奇异曲线。其次,构建了决策映射,对相同子带上的各SR系数的活性度和匹配度进行决策分析,输出决策值,通过决策值对图像进行匹配融合。最后,通过多尺度逆变换得到最终的融合图像。仿真实验表明:与当前图像融合算法相比,获得的融合图像具有更好的视觉效果,能有效图像突出目标信息,得到的图像具有更高的平均梯度和边缘评价因子;既突出了目标特征又保留平均强度信息,同时降低噪声影响。

The Image Fusion Algorithm Based on Multi-resolution Analysis Coupled with Approximate Sparse Representation

In order to better deal with the high dimensional features of image singularity,fusion image target feature and average intensity information, an image fusion algorithm based on multi-resolution analysis and approximate sparse representation was proposed. Firstly, the source images were multi-resolution analyzed, and the high frequency and low frequency information of the image are obtained; Then, the approximate sparse representation (Representation Sparse, SR) is designed, and the high frequency and low frequency information of the image are represented by the approximate SR coefficient, by using the absolute maximum selection technique to approximate the SR sparse transformation, the approximate coefficients of the low frequency sub band and the detail coefficients of the high frequency sub bands are obtained. The "anisotropic" function is constructed to approximate the singular curve with the least coefficient; thirdly, the decision mapping is constructed, and the SR coefficients of the same sub bands are analyzed and the matching degree is analyzed, image matching and fusion based on decision value; Finally, the final fusion image is obtained by multi-scale inverse transform. The simulation results show that compared with the current image fusion algorithm, the fusion image obtained by the proposed algorithm has better visual effect, The image can be effectively image to highlight the target information, the image has a higher average gradient and edge evaluation factor, which not only highlights the target feature and retains the average intensity information, while reducing the impact of noise.