Research On 3D Texture Mapping And Enhancement Of Real Scene

With the development of Smart Cities and Digital China,the creation of high fidelity and realistic 3D models of cities with textures has become increasingly important.As the most effective way to enhance the realism of 3D models,texture mapping has rapidly become a hot research topic in the fields of computer graphics,computer vision and photogrammetry.After years of development,texture mapping technology has basically been able to meet the needs of practical applications,but there are still many problems,such as:(1)the surface model obtained by the current model surface reconstruction algorithm has the problem of non-manifold,which can not provide direct use of the basic data for texture mapping,and requires non-manifold processing later.The current mesh network non-manifold processing algorithm has many problems,such as the existence of unmanageable non-manifold structure,will produce holes,repeat faces,repeat points and other network structure problems,and inefficient.(2)The direct results of the current urban 3D texture mapping algorithm have problems such as blurred textures,texture fragments and gaps,which affect the realism and readability of the textures and need to be further optimised.Therefore,it is important to study texture mapping and enhancement methods to improve the clarity and smoothness of textures.This thesis addresses the above-mentioned problems of current realistic 3D texture mapping and conducts research on texture mapping and texture realism enhancement methods,with the aim of improving texture quality and establishing high-fidelity and readable realistic 3D models.The main work of this thesis is as follows.(1)In terms of texture mapping,an improved method for non-manifold processing of mesh nets,a coarse screening method for visual images of sub-blocks,and a multi-parameter adaptive weighted method for high-quality texture mapping based on graph cut are proposed.Firstly,to address the above-mentioned problems of the current non-manifold processing algorithm,the method of open source library VCG and OpenMVS are analyzed in depth for their principles,advantages and disadvantages of detecting and processing non-manifold,and an improvement of the OpenMVS non-manifold processing method is implemented based on the VCG library.Secondly,to address the problem of inefficient visual image selection of sub-blocks,a coarse screening method of visual images of sub-blocks is proposed using the method of plane intersection to improve the efficiency.Finally,A multi-parameter adaptive weighted method for high quality texture mapping based on graph cut is proposed to address the problem that the current texture mapping algorithm cannot perfectly balance texture clarity and smoothness.The method is based on the traditional graph-cutting algorithm,which integrates the image and object areas of triangular facets,the lengths of edges and the normal angle between adjacent triangular facets to construct a multi-parameter adaptive weighted energy equation,and optimizes this energy equation by the graph-cutting algorithm to improve the quality of textures.(2)In terms of texture realism enhancement,using the idea of Poisson editing,the inter-block texture homogenisation is achieved,and a texture fragment removal method is proposed.Firstly,to address the problem of colour and luminance mutations in inter-block textures after the reconstruction of model by blocks,the principle of 2D Poisson image fusion is analyzed and applied to the homogenisation of inter-block textures of 3D models,so as to make the colour and luminance of inter-block textures consistent and enhance texture realism.Secondly,to address the problem of texture fragments in the current texture mapping results,a texture fragment removal method with forced replacement of texture images is proposed to enhance texture smoothness.The experiments show that the non-manifold processing method of mesh network in this thesis can significantly improve the effect and efficiency of non-manifold processing;the texture mapping method can improve the clarity of textures while reducing texture gaps;the texture realism enhancement method can effectively remove texture fragments and texture seam lines between blocks to enhance texture realism,and thus provide a higher quality and more realistic urban 3D model block for smart city and digital city construction.