Image Fusion Algorithm Based On PCNN And NSCT Transform

With the continuous development of image sensor technology, the technologyof image fusion as a branch of information merging technology, have gainedwidespread application in many areas. In order to get a fusion method with goodeffect and fast processing speed image, the basic theories and concepts of the existingdigital fusion technology are discussed and analyzed in the paper. Some domestic andabroad typical image fusion algorithm are researched and analysised based on theexsisting image fusion algorithm. the image fusion method Based on a combination ofnon-mining under Contourlet transform and adaptive PCNN is proposed according tofeatures of NonSubsampled Contourfet transform, and pulse coupled neuralnetwork.The main innovation of the paper work are as follows:1. Researched the use of multi-scale decomposition and reconstruction tools interms of image fusion,and the defects of multi-scale decomposition andreconstruction tools such as multi-scale wavelet transform and that usual ContourletTransform tool does not have the translation invariance which is easy to produceGibbs phenomenon,by choosing NSCT to transfer the source image frequencydomain, Direction information of the image can be digged, the high abramovichheterosexual og image can be expressed sparsely, Multi-directional and multi-scalefeatures of image can be reflected;as the translation invariance of NSCT transform,theGibbs phenomenon is overcome.the usual principle of Nonsubsampled Contourlettransform and PCNN is introduced.2. By comparing the performance of the fused image NSCT transform imagefusion method based on the resulting performance with different multi-scale imagefusion decomposition and reconstruction tools obtained, we can draw that the NSCTtransform based image fusion method is superior to the traditional generalconclusions.The fusion method based on a combination Nonsubsampled ContourletTransform and Adaptive PCNN are proposed. 3.On the issue of designing the image fusion rules, using different rules to thesouse image’s low-frequency and high-frequency sub-band subband decompositioncoefficients by nonsubsampled Contourlet transform. For low-frequency sub-bandfeature images and bandpass directional subband features were selected based on theamount of image regions characterized as low frequency sub-band coefficientselection rules and selection rules based on PCNN as bandpass directional subbandcoefficients. And for the traditional PCNN model in the intensity image fusion linksare based on a fixed value of the shortcomings, the definition of image pixels basedon regional characteristics as PCNN model neuron connection strength of selectionrules, so that the neuron can PCNN model according to the different regionalcharacteristics of image pixels adaptively select the link strength.4.Many traditional image fusion method based on PCNN models are usingimage pixel gray value as an external input neurons PCNN.However, according to thehuman visual system is sensitive to the edge of the image details and characteristics ofthe image shows the information, the features of image are often not reflected by asingle pixel, only a single pixel in the gray value to an external input neurons PCNNis not enough. To overcome the defects of traditional methods, the features of imagebased on different types to be fusion image in used in the paper,by selecting differentarea characteristics as PCNN neurons as external input method, and according to thenumber of neurons firing to choose the size of the fused image NSCT factor.By using different multi-scale decomposition and reconstruction tools for imagefusion of different image sensor obtained with this method fused image obtained bycomparing obtained demonstrate the effectiveness and superiority of this method.Theexperimental results show that the fusion method proposed in this paper in terms ofhuman visual effect or objective evaluation criteria are superior to the fusion methodbased on traditional multiple degrees decomposition and reconstruction tools.