Multi-focus Image Fusion Algorithm Based On Focus Measurement

In recent years,imaging devices have been mass-produced and popularized,and image data collection has become easier.Image data contains more information,which is more intuitive and more acceptable to humans compared to other types of information.As a research hotspot in the field of image processing,image fusion is the process of taking a series of images with complementary information as an operation object and integrating these feature information to obtain an image.The fused image contains more abundant and accurate information,and the visual effect is better,which is more conducive to subsequent image processing.Multi-focus image fusion is an important part in the field of image fusion.A multi-focus image refers to the image with different sharpness due to different distances between different objects and lenses when using optical imaging system to shoot three-dimensional scene.Multi-focus image fusion refers to the process of fusing two or more multi-focus images to produce an all-in-focus image.It is widely used in the fields of target recognition,military operations and other applications,and has important significance.This paper mainly studies the multi-focus image fusion algorithm based on focus measurement.Firstly,this paper introduces the related knowledge of multi-focus image fusion,including classification and hierarchical division of image fusion,current research status and algorithm classification,fusion image evaluation and so on.Then,combining the traditional transform domain method and the spatial domain method in the field of multi-focus image fusion,this paper proposes an image fusion method based on focus measurement(MST-FM).A multi-focus image fusion algorithm based on multi-scale transform is proposed.Firstly,the initial multi-focus images are decomposed by nonsubsampled shearlet transform;then different fusion rules are adopted in the high-coefficients and the low-coefficients respectively;finally,inverse transform is performed to reconstruct an initial fused image.We use the similarity measurement between the original fused image and the source images as the focus measurement,and the decision map is obtained.The final fusion image is formed according to the decision map.Through the comparison of the subjective and objective experimental effects with the classical algorithm of the field of multi-focus image fusion,it is illustrated that MST-FM combines the advantages of the transform domain method and the spatial domain method,it effectively improves the fusion effect.Moreover,an emerging multi-focus image fusion method based on full convolutional network(FCN-FM)is proposed.The algorithm applies the deep learning algorithm which is excellent in the field of computer vision to multi-focus image fusion,and uses the convolutional neural network to measure the focus of the source images.Each source image is used as the input of the network,and the output is the focus map of the corresponding image.The focus map is a pixel-level clarity measurement of source image.An all-in-focus image can be combined by finding relatively clear pixels.The network in this paper can combine multi-scale information to improve the quality of fusion,and the full convolution network can process images of any size.FCN-FM is effective and easy to understand.The experimental results show that compared with the traditional method,FCN-FM avoids blocking effect and artificial effect,and has strong robustness.And the fusion effect of the FCN-FM is better than the comparison algorithm including MST-FM from the perspective of subjective and objective evaluation,achieves the purpose of this experiment.