Research On Feature Matching Algorithm Of Light Field Image Based On Local Invariance Analysis

Feature detection,image matching and is described as the basis for image processing is critical to the development of many computer vision technology and applications.Traditional 2D images can only focus on a certain fixed depth in the scene,and targets at other depths in the scene are blurred.Therefore,the traditional 2D image feature detection and matching cannot accurately and comprehensively detect the features of objects at various depths in the scene.The 4D light field image saves 2D position information and 2D angle information in the scene at the same time,and can obtain images focused at different depths through digital refocusing operations.Therefore,this paper mainly studies the feature detection and matching methods for 4D light field images.The local invariant feature of image is very important for image feature detection and description.The local invariance feature of an image refers to the image feature that can still be detected accurately when the scale,rotation,illumination and other changes occur,and then the feature description is used to represent the relationship between the current point and its surrounding points and so on.Features based on local invariance analysis are more stable and independent.Therefore,this topic proposes a 4D light field image feature detection and matching method based on local invariant feature analysis.The main research contents and innovations of this topic are as follows:(1)In view of the characteristics of the 4D field image that can be focused at different depths through the digital refocusing operation,but the 4D field image has a lot of redundant data.This topic reduces the 4D light field image into a 3D focal stack image.This process not only effectively obtains images focused at different depths,but also greatly reduces redundant data and simplifies calculations.(2)Considering the scale invariance in the analysis of local invariance features,this topic proposes to introduce a Gaussian scale space in a 3D focal stack image to construct a 4D scale-depth space with both scale invariance and depth characteristics.Furthermore,the Harris feature detection operator with rotation invariance is used for feature detection in the 4D scale-depth space.In addition,in order to eliminate the influence of illumination,the feature descriptor was normalized to make the detected feature have illumination invariance.Therefore,the 4D light field image features detected by this subject simultaneously satisfy scale invariance,rotation invariance,illumination invariance,and depth characteristics.(3)In order to verify the correctness and effectiveness of the multi-scale Harris feature detection method of the light field image proposed in this subject,this subject uses a homography matrix to verify the image matching results and calculate the ratio of the number of mismatched pairs to the total number of matched pairs.The experiments are compared with SIFT feature detection and matching methods on traditional 2D images and LIEF feature detection and matching methods for light field images proposed by Stanford.The results of simulation experiments prove the effectiveness and accuracy of the multi-scale Harris feature detection and matching method proposed in this paper.(4)For the fusion of multi-sensor images,based on the multi-channel,multi-scale characteristics of the human visual system,this subject proposes contrast pyramid decomposition and directional filter banks decomposition for multi-sensor images for contrast and directional feature detection.Then the whale optimization algorithm is proposed to optimize the adaptive fusion coefficient,and finally complete the multi-sensor image fusion.This project is compared with four classical fusion methods,and the experimental results are analyzed and summarized using six numerical indicators.Simulation experiments show the accuracy and efficiency of the directional feature detection and contrast fusion methods proposed in this subject.