深度学习表达的图像特征训练与融合算法

Image feature training and fusion algorithm based on deep learning representation

图像的融合可以展示不同状态下拍摄出的照片的形态,综合它们各自的信息得到一幅信息量丰富的图像.利用深度学习的方法可对图像进行特征的提取与融合.首先将几幅原始图像采用滑动窗技术进行分块,然后再组合形成联合矩阵,用深度学习的一种稀疏自编码器模型训练出权值矩阵和参数矩阵,通过反馈对参数进行微调,从而得到几幅原始图像对应的特征,最后采用最大化选择的方法融合原始图像.实验结果表明,本文算法具有良好的融合效果.

The forms of photos taken with different states can be shown by image fusion,combined with their respective information to get an image of the rich informations.In this paper,the deep learning method is used to extract and fuse the features of the image.Firstly,several original images are segmented by sliding window technique,and then combined to form a joint matrix.A sparse self-encoder model of depth learning is used to train the weight matrix and parameter matrix,and finely tune the parameters by feedback.The corresponding features of several original images have been gotten,these features are automatically learned through the sparse auto encoder.Finally,the original images are fused by the method of maximum selection.Experimental results show that the algorithm proposed in this paper has a good fusion effect.