Image Fusion Algorithm Research Based On Visual IoT

In recent years, as the development of wireless sensor network with perceptivity,computing ability and communication ability, smart visual Internet of Things has been widelyused in various fields. It can satisfy the requirement of collection and compression ofmultimedia information like images, audios and videos in practical applications, which has abroad application prospect.This paper focuses on image fusion algorithm in visual Internet of Things, introduces thethree levels of image fusion and analyses two main methods of pixel level fusion: fusion inthe spatial domain and fusion in transform domain. In research of image fusion, fusion rulesare very important for the fusion result. Based on the previous research results, this paperproposed some new image fusion algorithms with corresponding fusion operators.First, propose an image fusion algorithm based on image edge information. Because ofthe edge information of clear parts in the images to be fused are more other parts of theimages, this paper combines the fusion algorithm based on pixels and fusion algorithm basedon blocks. First, using adaptive threshold to extract the clear parts of the fusion image, andthen using the method of weighted mean to fuse the critical zone to get the final result.Simulation results show that this algorithm could reserve more clear parts of the images to befused, which has better effect than traditional fusion algorithm in spatial domain.Secondly, propose an image fusion algorithm based on improved Laplacian Pyramid. Inthe traditional Laplacian Pyramid, it’s unable to avoid the noise in the reconstructionprocedure. In view of that, this paper uses an improved Laplacian Pyramid to reconstruct theimages. At the same time, in view of the different feature of the top level image and otherlevel images, we use different fusion operators to deal with. The experiment result indicatesthat this algorithm has a better effect than traditional ones, and noise in the fusion image issignificantly lower than the fusion image using traditional Laplacian Pyramid algorithm.Thirdly, propose an image fusion algorithm based on Wavelet Transform. Selection offusion operator often determines the fusion quality in the fusion algorithm based on WaveletTransform. Due to the approximation properties of low frequency image, the percentage ofenergy is large, we use the method based on neighborhood energy to deal with the lowfrequency part. Meanwhile, variance could reflect the characteristics of high frequency in theimage edge, so we use the method based on neighborhood variance to fuse the high frequencypart. Simulation results show that, this algorithm could achieve good fusion result, and verifythe size of neighborhood’s impact on the fusion result.