Research On 3D Image Processing And Mosaic Technology

Since 3D images have important applications in reverse engineering,pattern recognition,artificial intelligence,etc.,3D image processing technology has become a hot research topic in the field of 3D images.This paper focuses on the key technologies of 3D image mosaic,surface reconstruction,model simplification,etc.,and studies the 2D image filtering and splicing.The main research results are as follows:1.For the two-dimensional image contaminated by noise,the MRF(Markov Field)image filtering algorithm is improved to filter the noise image.The MAP-MRF(Maximum a Posteriori-Markov Field)system is established.Using the equivalence of MRF and Gibbs,the relationship between each point of the image and the global is obtained,and the Iterated Conditional Model(ICM)is improved.The algorithm in this paper can effectively filter out the noise to get the original image.2.A splicing algorithm based on spectrum and spatial feature matching is proposed for two-dimensional image mosaic scenes with inconspicuous features.Based on the invariance of the rotation center of the spectrogram,the amount of rotation of the image around any point is converted into the amount of rotation around the center of the spectrogram.Using polar coordinate system and Cartesian coordinate system conversion,the rotation amount around the center of the spectrogram is converted into one-dimensional translation,the rotation angle is obtained by SAD(Sum of Absolute Difference)algorithm,and the translation matrix is obtained by SAD algorithm.The algorithm in this paper is suitable for splicing scenes with inconspicuous features,and is also suitable for splicing scenes with obvious feature points,and the registration accuracy and speed are high.3.A stitching algorithm based on 3D image features is proposed.The feature point set is extracted according to the three-dimensional image normal vector information,and the histogram feature vector is established for description,and the initial matching point pair is obtained.According to the distance invariance of the rigid transformation,the RANSAC(Random Sample Consensus)algorithm is used to remove the error point pairs and obtain the initial transformation matrix.Fine registration using the ICP algorithm(Iterative Closest Points).The algorithm provides a fast and accurate coarse registration algorithm,which provides an accurate initial position for the ICP algorithm and improves the overall registration accuracy.4.A clipping algorithm is proposed for the Poisson reconstruction algorithm applied to the error patch obtained from the non-closed 3D image.Combining the relationship between the model vertices and the three-dimensional image points and the distribution law of the surface patches of the model,the confidence of the model vertices is established,and the triangular patches connected with the vertices whose confidence is less than the threshold are deleted,so as to obtain the three-dimensional model matching the original images.The mesh model obtained by the algorithm has high matching degree with the original image,and the cutting size can be controlled by adjusting the threshold,which improves the practicability of the algorithm.5.A simplification algorithm based on octree is proposed.The octal tree is used to divide the mesh model to provide a new curvature estimation method for each voxel patch cluster.The voxels whose curvature is less than the threshold are deleted,and the simplified octree is obtained.The simplified model is obtained by using the moving cube algorithm(Marching Cubes).The algorithm in this paper preserves the characteristics of the model while reducing the number of patches.