The Research On Multi-focus Image Fusion Algorithm Based On Deep Learning

Due to the limitation of the depth of field of the optical element,it is difficult to capture all the clear targets in an image.Only the target at a specific position from the camera can be clearly focused.The target before or after the focus plane will lose focus and become blurred.In order to solve this problem,a multi-focus image fusion algorithm is proposed,the purpose of which is to fuse images with different depths of field into an all-focus image,so as to obtain a more comprehensive and reliable scene description.At present,multi-focus image technology is widely used in image enhancement,digital imaging and other fields.In recent years,a variety of multi-focus image fusion algorithms have been proposed.According to different image fusion methods,multi-focus image fusion methods include transform domain-based and spatial domain-based methods.These methods have good performance in extracting and expressing image details,but the disadvantage is that the manual design of activity measurement and fusion rules is difficult,and there are many factors that cannot be fully taken into account.Since deep learning has strong feature extraction and data representation capabilities,it is outstanding in image processing and machine vision tasks,so the use of deep learning to solve multi-focus image fusion problems has also become a topic of great concern.The main research contents of this article are as follows:(1)To further improve the fusion quality of multi-focus images,a multi-focus image fusion algorithm for supervised learning under a fully convolutional neural network is proposed.The algorithm aims to use neural networks to learn the complementary relationship between different focus areas of the source image,that is,select different focus positions in the source image to synthesize a globally clear image.The algorithm constructs focused images as training data,and the network uses dense connections and 1 * 1 convolution to improve the network's understanding and efficiency.The result of the experiments shows that the algorithm is superior to other comparison algorithms in both subjective evaluations based on visual observation and objective evaluation based on evaluation indicators,and the image fusion quality is further improved.(2)A multi-focus image fusion algorithm based on generative adversarial network is proposed.The generative network introduces residual block structure and dense connections;in order to reduce the artifacts in the fused image and ensure the quality of the image fusion,a more detailed cascade loss function with perceptual loss is proposed.The experimental results show that,compared with other comparison methods,the proposed method has achieved good results in both objective evaluation and subjective evaluation.