面向资源受限终端全景纹理数据压缩算法的研究

发布时间：2018-04-28 22:02

本文选题：资源受限终端 + 全景图　；参考：《太原理工大学》2017年硕士论文

【摘要】：三维场景可视化技术在数字地球、数字旅游和3D GIS等领域应用广泛,其中纹理图像的再现效果直接影响用户体验,是体现三维场景真实性的关键因素。三维场景中景物的立面、天空、大地等都需要用纹理图像来增加场景的逼真度。对于手机、平板这类分辨率低、内存量小、流量和电量等资源受限的移动终端,三维场景可视化中纹理图像的渲染一般需要大量的纹理图像数据,大量数据的传输会消耗巨大的网络流量和电量,阻碍着三维场景可视化技术在移动终端上的应用和发展。以降低数据量为主的纹理图像渲染方法中,Mip-Map和逆过程式语法纹理构建两种方法脱颖而出。Mip-Map,当观察者视点与景物之间的距离改变时,景物立面纹理图像会发生低分辨率与高分辨率相互切换的过程,同一景物立面的不同分辨率纹理图像多次传输与加载,这个过程会增加从服务器端到资源受限终端的纹理图像的数据量。使用Mip-Map需要有视点到景物的深度信息,只适用于基于图形学的建模方式。逆过程式语法纹理构建方法在渲染时,由于用户需要通过逆过程式建模将单元纹理重构还原真实纹理,所以会增加渲染终端的计算量。基于此,本文在对比传统三维场景建模方法的基础上,采用三维几何模型数据较少的全景建模方法,对数据占有量最大的纹理图像进行压缩。提出一种面向资源受限终端基于信息熵的全景纹理图像数据压缩算法,实现多级纹理细节的立方体全景纹理图像的渲染:根据人眼视觉对信息的选择特点,对场景中关注度高的区域,使用高分辨率纹理图像渲染,对场景关注度低的区域,使用低分辨率纹理图像渲染,以达到减少所需的纹理图像数据和降低移动终端资源消耗的目的。使用图像信息熵表示场景各区域的复杂度,按照信息熵的大小对场景各区域分级,不同级别的区域采用不同级别的纹理图像分辨率进行渲染,使之达到全景图像的同层次多级细节(LOD-in-IBR)。对于图像缩放算法所产生的块效应,本文采用双三次迭代插值算法进行修正。实验从压缩率、实时性、均方误差(MSE)和峰值信噪比(PSNR)和特征相似度(FSIMc)衡量图像质量以评价纹理图像压缩质量,证明本文提出的方法在压缩纹理图像数据30%的情况下,所生成的全景图像的视觉质量仍然不低于原有全景图像的视觉质量。
[Abstract]:3D scene visualization technology is widely used in the fields of digital earth, digital tourism and 3D GIS, in which the reappearance effect of texture image directly affects the user experience and is the key factor to reflect the reality of the 3D scene. The scene, sky and earth in the three-dimensional scene need to be used to increase the fidelity of the scene. As a mobile terminal, such as low resolution, small memory, and limited resources, such as low memory, flow and electricity, the rendering of texture images in 3D scene visualization usually requires a large number of texture image data. The transmission of large amounts of data will consume huge network traffic and electricity, which hinders the application of 3D scene visualization technology on mobile terminals. And development. In the texture image rendering method based on data reduction, the Mip-Map and the reverse program syntax texture build two methods to stand out.Mip-Map. When the distance between the view point and the scene of the observer is changed, the scene texture image of the scene will switch between the low resolution and the high resolution rate, and the same scene is different from the same scene. Resolution texture images are transmitted and loaded multiple times. This process increases the amount of texture images from the server to the resource constrained terminal. The use of Mip-Map needs to have the depth of view to the scene, only for graphics based modeling. The inverse process syntax texture construction method is required by the user to pass through the rendering. In this paper, based on the comparison of the traditional 3D scene modeling method, this paper uses a panoramic modeling method with less data of 3D geometry model to compress the most large data possession of the texture image. The source limited terminal based on the information entropy based panoramic texture image data compression algorithm to achieve the multi-stage texture details of the cube panoramic texture image rendering: according to the human vision of the information selection characteristics, the area of high attention to the scene, using high resolution texture image rendering, low area of attention to the scene, the use of low resolution Texture image rendering, in order to reduce the desired texture image data and reduce the consumption of mobile terminal resources. Using image information entropy to represent the complexity of the various regions of the scene, according to the size of information entropy, each area of the scene is graded, and different levels of the texture image resolution is used to make it complete. The same hierarchical detail (LOD-in-IBR) in the image. For the block effect produced by the image scaling algorithm, this paper uses the double three iteration interpolation algorithm to modify the image. The experiment is proved from compression rate, real-time, mean square error (MSE) and peak signal to noise ratio (PSNR) and feature similarity (FSIMc) to evaluate the quality of image compression. In the case of compressed texture image data 30%, the visual quality of the generated panoramic image is still not less than the visual quality of the original panoramic image.

【学位授予单位】：太原理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】