Image Depth Calculation And Application Research Based On Markov Random Field

Image depth refers to the depth of each point in the image(corresponding to each pixel on the image)corresponding to the point in the scene,specifically the distance between each pixel and the camera in the image.By analyzing and recognizing the preprocessed two-dimensional images,we can estimate the relative distances of the objects in the scene in the real world.Image depth information is widely used in various fields.It is a hot issue of computer experts and many researchers in recent years.It has broad commercial prospects.In recent years,with the rapid development of image depth information technology,images have been applied in many fields.Accurate estimation of depth information from images is the key point of three-dimensional reconstruction.It can better understand the image scene,and it is also a basic problem in the field of computer vision.This will effectively promote the application of computer vision in many fields,such as Ann.Full monitoring,intelligent robots,medical images,2D to 3D,etc.For several commonly used image depth information extraction methods,the MRF model algorithm is the typical one.In view of the MRF model image depth extraction method,not only considering the local features of the image,but also considering the global features of the image,it can extract image features more comprehensively,so that it can quickly choose appropriate image features from the sub-data sets for depth extraction.This method has certain practical value and theoretical significance.This paper focuses on the MRF model of image depth acquisition and non-parametric learning.Firstly,the basic concepts of image theory are introduced,and the common algorithms of image depth extraction are studied and analyzed.Then,the absolute depth feature,relative depth feature and position correlation depth feature of image are introduced.On this basis,the MRF model is introduced comprehensively: not only texture feature is considered.Direction change also takes into account the spatial distribution of structural texture features.Finally,combined with gradient sample data sets,KNN algorithm is used to train the depth values of samples,and non-parametric learning is used to estimate depth maps.Specific research results are as follows:(1)Local group Potential Energy Modeling Using MRF.This paper elaborates three characteristics of image depth: absolute depth feature,relative depth feature and position relationship feature,so as to understand image features more comprehensively;MRF model is established by considering the direction and spatial distribution of image texture,which has fewer parameters and is more convenient to process.MRF considers the local pixels of the image.A certain pixel is only related to the adjacent pixels.According to the potential energy of the local group,the global pixels are considered.(2)Non-parametric learning is used to estimate image depth.By transferring the image features extracted from RGB-D database,KNN nearest neighbor method is used to train the depth gradient of the sample,and MRF model algorithm is used to synthesize the depth gradient field,which takes into account both the local features of the image and the global features of the image.Finally,the gradient field is combined with Poisson surface reconstruction to extract the depth map of the image and estimate the depth of the image.Reconstruction of dimension.The experimental results show that,compared with the traditional algorithm,the depth map estimated by this algorithm has higher accuracy,smaller error and wider practical range.Combined with the ideas of this paper,the above research results are simulated and the relevant depth maps are generated.The comparison is compared with the relative error and root mean square error in the classic Make3 D algorithm.The Saxena algorithm and the Zhuo algorithm are used to verify the results.The root mean square gradient error and the processing time consumption of the image are validated and analyzed,and the effectiveness of the experiment is verified more comprehensively.