Research On Pixel-Level Fusion Algorithms For Multi-focus Image

Image fusion is an important branch of information fusion.It is closely related to sensors,signal processing,image processing,artificial intelligence and other disciplines.It is an effective way to obtain image information accurately.Multi-focus image fusion is a research focus in the field of multi-source image fusion,which overcomes the problems of single image in spectral,geometric,spatial resolution and other aspects.Among them,pixel-level fusion information is less lost and more detailed information is acquired,which has become the mainstream of multi-focus image fusion research.However,the pixel-level fusion requires high accuracy of registration,and requires a large amount of image details to be processed during the fusion process,resulting in poor real-time performance.This paper deeply studies the key issues and links in the process of pixel-level image fusion.The specific work is as follows:Aiming at the problem of inaccurate and slow operation of the initial matching point in the image registration process,a feature matching method based on parallax constraint and cluster analysis is proposed.Firstly Harris corner detection and Normalized Cross Correlation(NCC)function are applied to the source image.The rough matching is performed,then the K-means clustering is performed on the rough matching results by using the distance feature vector and the direction feature vector respectively.Finally,the Random Sampling Consensus(RANSAC)is used for the second optimization.This paper introduces the idea of cluster analysis to improve the precision of matching pairs and reduce the precision matching time.The experimental results show that the registration algorithm based on parallax constraint and cluster analysis can eliminate most of the false matching points,improve the accuracy of the traditional image matching algorithm,and shorten the running time.In order to solve the problem of relatively low fusion accuracy in traditional multi-scale image fusion algorithms based on multi-scale transforms,an image fusion method based on Dual Tree Contourlet Transform(DTCT)is proposed.Firstly,the multi-focus image is separately decomposed into a Contourlet decomposition of the duality tree to obtain a low-pass.Bands and band-pass subbands,followed by the fusion rule with the largest regional standard deviation for the low-pass subbands,the largest absolute fusion rule for the bandpass subbands,the processing of the coefficients,and the reconstruction of the processed coefficients.The fused image is obtained.The experimental results show that the multi-focus image fusion method based on dual tree Contourlet transform is superior to the traditional multi-scale image fusion method based on multi-scale transform.The fusion accuracy is high,and the image details remain intact.The proposed algorithm is superior to the contrast algorithm in most of the indicators,and it is similar to the advanced algorithms.