论文部分内容阅读
基于多聚焦图像融合中存在的低频信息易产生缺失的现象进行分析,提出一种基于深度神经网络模型的低频子带融合策略,并结合小波核滤波器及针对高频子带的融合策略,给出多聚焦图像融合方法。该方法利用自动编码器提取低频子带特征,利用网络隐层中的权值信息选择低频子带分量。采用3组聚焦不同的自然图像及1组医学图像进行算法测试,并与传统的低频子带融合策略进行对比,同时比较基于轮廓波变换的多聚焦图像融合方法、基于非下采样轮廓波变换的多聚焦图像融合方法。试验结果表明:其中一组图像采用深度神经网络模型的策略所得到的融合结果的边缘融合指标值能够达到0.802 7,优于其余比较方法的0.761 4、0.722 7和0.716 4,从而证实基于深度神经网络模型的融合策略的有效性。
Based on the analysis of low-frequency information prone to loss in multi-focus image fusion, a low-frequency subband fusion strategy based on deep neural network model is proposed. Combining with the wavelet kernel filter and the fusion strategy for high frequency subband, Out of multi-focus image fusion method. The method uses the automatic encoder to extract low frequency subband features, and uses the weight information in the hidden layer of the network to select low frequency subband components. Three groups of different natural images and one group of medical images are used to test the algorithm. Compared with the traditional low frequency subband fusion strategy, the multifocus image fusion method based on contourlet transform is compared. Based on the nonsubsampled contourlet transform Multifocus Image Fusion Method. The experimental results show that the edge fusion index of the fusion results obtained by using the neural network model of one group of images can reach 0.802 7, which is better than the other comparison methods of 0.761 4, 0.722 7 and 0.716 4, Validity of neural network model fusion strategy.