Research On Infrared Image And Visible Image Fusion Based On Multi-wavelet Transform

Different kinds of sensors have different imaging principles, they provide different formsof image datas. In order to take full advantage of the information provided by each image,reduce storage space for these large numbers of images, the concept of image fusiontechniques comes into being.Image fusion technology can eliminate redundant data amongmulti-sensor, improve information utilization, formation of the overall, clear and accurateunderstanding of the objective. The technology has been widely applied to medical imaging,remote sensing imaging, public safety, industrial control, transportation regulation and otheraspects.This paper studies the fusion algorithm of visible and infrared images, the image fusedcan be fully reflect the target object in sheltered or in low-light conditions. Image fusion canbe divided into three parts, include image decomposition, integration and image reconstructioncoefficient, this paper improves the image decomposition-reconstruction method and fusionrules, the new algorithm is used in the fusion of infrared images and visible images, the maincontents include the following aspects:First this paper studies the two-dimensional image decomposition and reconstructionbased on multi-resolution analysis, focuses on wavelet transform and multi-wavelet transform.The traditional image decomposition methods such as Laplace transform and HPF high-passfiltering method can lead to data redundancy, to avoid this problem this paper selects GHMMulti-wavelet and the famous Mallat wavelet algorithm for image decomposition andreconstruction, GHM multi-wavelet has excellent characteristics, Mallat algorithm canachieve rapid decomposition and reconstruction and does not produce data redundancy,thedecomposition method can decompose the source image into different frequencysub-bands,maintain the frequency characteristics of the image.Secondly, this paper studies on the fusion rules deeply. Fusion rule based on the regionalcharacteristics is computing eigenvalues within a region, which avoiding the blocking effectcaused by calculating the value of a single pixel. After multi-resolution decomposition,theimage is decomposed into different frequency subbands. The low frequency section containsthe main energy, the pixel value difference is negligible, generally take the weighted averaging algorithm; High-frequency part of image include detail features, usually using gradient rulesalgorithm.This paper improves high frequency gradient fusion rules，put forward to gradientvalue comparison method, combine the high and low frequency fusion rule to form a completefusion algorithm,can improve the quality of image fusion effectively.Finally, this paper conductes experiments. After simulating by matlab and calculating theperformance indicators such as information entropy, standard deviation and averagevalue,comparing with wavelet transform method, we can see that using the improvedalgorithm can improve the fusion quality, the new algorithm contributes to further use of theimage.