Research On Image Fusion Method Based On Multi-scale Transform

Image fusion can combine different images of the same scene into a fusion image through a specific algorithm,which makes the quality of the fusion image higher,the information in it richer and information redundancy of it lower.Image fusion technology has now been applied in remote sensing,medical,military,security and other fields.At present,although there are many scholars in the field of image fusion has been studied in-depth,however,because of the useful information in the image extraction complexity and different types of sensors collected information specificity,image fusion technology has not yet reached the effect that people expected.This paper firstly introduces the research background of image fusion,the current situation of image fusion research and the existing fusion methods.The common image fusion method based on multi-scale transform is introduced in detail,and the advantages and disadvantages of these fusion methods are summarized and analyzed.Then,this paper investigates how to improve the quality of image fusion by improving the multi-scale transformation tool and improving the existing fusion rules.Then,based on the multi-scale transformation of the image,a fusion method is proposed to improve the accuracy of apple quality detection.And the validity of the proposed fusion method is verified by experiments.Finally,we study how to improve the quality of image fusion by improving the multi-scale transformation tool and improving the fusion rule of multi-scale coefficients.For the proposed fusion method,this paper introduces the theoretical basis,the fusion process and the implementation details of the fusion method in detail.A summary analysis is made from the subjective feelings and objective results through the experiment of the all proposed methods.The main contents of this paper are as follows:(1)This paper presents a method to improve the accuracy of fruit quality detection by using image fusion,which is based on the problem that the existing fruit quality detection is based on single type fruit image.The fused image is obtained by fusing the infrared image and visible image of the fruit.In this way,the fusion image not only contains the visible image of the fruit epidermis information,but also contains the infrared image of the fruit subcutaneous injury information.The accuracy of the detection results is improved by using fusion image.(2)Through the deep study of image fusion based on multi-scale analysis,a quaternion wavelet transform is proposed for quaternion wavelet transform only to satisfy the approximate translation invariance.Compared with other multi-scale transformation tool,non-subsampled quaternion wavelet transform has the feature of translation invariance,and can provide rich scale information,the amplitude and phase information,Amplitude can better represent the image texture features,and the phase can provide richer geometric information.The final fusion results acquired from the fusion method based on the quaternions wavelet transform can not only have outstanding objective data performance,but also get the best results of subjective effects.(3)The traditional image fusion algorithm only according to a single feature of the image to fuse image,which can not effectively extract the information to be integrated in the image.For this problem,an image fusion method based on synthetic feature is proposed.The use of a number of different features from a number of angles to describe the nature of the image,by assigning different weights for each feature to get a comprehensive feature.This comprehensive feature effectively avoids the single feature caused by incomplete extraction of information,hence,we construct the fusion algorithm,which can get better fusion results,and enhance the universality and robustness of the fusion algorithm.The effectiveness of the algorithm is verified by different types of image fusion experiments.(4)In the selected image fusion method,the selection decision graph calculated from the multi-scale transform coefficients will show some voids and noises due to the influence of noise.The traditional solution is to use morphological operations to eliminate this part of the interference,and morphological methods can not tell whether the void in the decision map is caused by noise interference or empty in the original image.In order to solve the problem,guided filtering is introduced to replace the morphological operation.The guided filtering can guide the image to be fused and effectively filter out the noise interference point without losing information.