Research Of Image Fusion Technology Using Multi-Scale Transformation And Neural Network Model

Benefit from the rapid development of computer vision technology,as an important branch of computer vision,image fusion technology has been attached great importance by researchers from all over the world in the field of digital image processing.When there are multiple target objects in the same scene,it's hard to get a picture with all the targets are clear at the same time,so the meaning of image fusion is to make full use of the complementary advantages of multi-source image information,which provide us a simple and efficient way to overcome the imaging defects in optical lenses,thus,we can fuse multiple images into a new image that contains all the valid information.With the continuous innovation of image fusion technology,the output of this technology is more conducive to the understanding of image information by human visual system,image fusion technology has been utilized for application in military,medical,remote sensing mapping,industrial testing and some other research fields.So it is of great significance to improve the performance of imaging system when make a deep research on image fusion technology.In the current mainstream image fusion algorithms,the algorithms which based on multi-scale transformation(MST-based)are undoubtedly the research hotspot in the field of image fusion.Non-subsampled contourlet transform(NSCT),as one of the MST-based method,it can effectively express the sparse characteristics of images,it inherits the time-frequency characteristics of traditional wavelet transform,and also realizes multi-direction characteristic of image transformation at the same time.Considering that NSCT causes image brightness and color distortion,which caused by the ignorance of the important spatial consistency of image,a hybrid l0-l1 sparse layer decomposition,as a newly developed image decomposition algorithm,has an effective capacity to overcome the defects of NSCT.Pulse coupled neural network(PCNN)is applied to the field of image fusion because it mimics the signal processing and transmission mechanism of the mammalian visual system,which saves the trouble of traditional artificial neural network needing a large amount of data for training.In recent years,with the introduction of the concept of deep learning,a convolutional neural network(CNN)based method for depth of image feature extraction can help to obtain a more complete and clear image.In this paper,some research of image fusion algorithms,which combines the MST-based methods and neural network model are proposed for multi-focus image fusion,infrared and visible fusion,and a large number of experiments were conducted to verify the effectiveness of the proposed algorithms,and the experimental results were evaluated from the perspective of subjective vision and objective index quantification.The experimental results show that the proposed algorith,m have a good effect in the field of image fusion.The main contents are as follows:1.Briefly introduce two principles of image decomposition model:NSCT based and hybrid l0-l1 sparse layer decomposition;2.Aiming at multi-focus image fusion,consider that the PCNN model can select the high and low frequency component coefficients of images through its special neuron ignition mechanism,combined with NSCT's property such as multi-scale,multi-direction,anisotropy and translational invariance,an image fusion method named multi-focus image fusion based on phase congruency motivate PCNN in NSCT domain is proposed.This algorithm decomposed the source image in multiple scales through NSCT,and developed a high and low frequency fusion strategy based on the image spatial frequency and phase consistency characteristics to detect the focusing region of the image.Combined with the PCNN ignition characteristics,the image gradient energy and low level non-deformation properties were used to effectively improve the multi-focusing image fusion effect;3.For multi-focus image fusion,due to the cross bilateral filtering(CBF)has a superior performance on image edge-preserving,an image fusion method named multi-focus image fusion using cross bilateral filter in NSCT domain is proposed,the algorithm firstly using NSCT to decompose the source image into high frequency subband and low-frequency subband,and through the calculation of the low frequency subband pixels of the Sum-Modified-Laplace to determine the focus area,at the same time using cross bilateral filter for high frequency subband decomposition again,which can help to achieve a superior fusion effect;4.As for infrared and visible image fusion,aiming at the unique imaging mechanism of infrared and visible images,an image fusion method named infrared and visible image fusion based on CNN model and saliency detection via hybrid l0-l1 layer decomposition is proposed,firstly,the source image is sparse decomposed by using the image sparse feature,then the infrared image significance detection is used as the basic layer component fusion rule and the depth feature extraction mechanism of convolutional neural network is used as the fusion strategy of detail components,and this kind of algorithm effectively integrates the thermal target information in the infrared image and the rich spectral information in the visible image.