【作者】 李雪松；

【导师】 王世刚；

【作者基本信息】 吉林大学 ， 信号与信息处理， 2015， 硕士

【摘要】 组合成像技术是一种自动立体技术，具有全彩色，连续视角等特点，能够逼真的再现3D场景。由于采集到的图像数据量大，有效的压缩以便于存储传输是十分必要的。对组合成像技术采集到的一幅静止图像目前常用的表示方式有三种：直接采集到的立体元图像阵列、子图像阵列、光线空间图像阵列，后两种图像阵列都可由立体元图像阵列转换得到。本文叙述了一种新的表示方式，这里称作“拼接图像阵列”，也是由立体元图像阵列转换得到的。分析了使用透镜阵列和采集面板这种采集方式采集到的一个立体元图像阵列中相邻立体元图像的光束重叠状况，光束重叠状况是影响图像间相似的原因之一，分析得出相邻立体元图像间光束重叠部分比例随着物体深度的增大越大，反映出相邻立体元图像间的相似性可能会增强；相邻子图像的光束重叠比例随着物体深度的增大而减小，反映出相邻子图像的相似性可能会减弱。对组合成像技术采集到的一幅静止图像的压缩，因对待图像的角度不同，可能有不同的压缩方法。可以把直接采集到的立体元图像阵列当作一个整体，作为一幅具有周期重复性纹理的普通图像；也可视作前面四种表示方式下的图像阵列。对图像阵列的压缩，主要有几种方法：使用针对性变换、用扫描曲线扫描成一维（1D）伪视频流、使用二维（2D）空间预测结构、压缩感知等其它方法。扫描成1D伪视频流相当于1D空间预测结构，只利用了图像阵列中图像在一个方向（水平或垂直）上的相关性，因此效果不是很好；二维空间预测结构能同时利用图像阵列中图像在水平和垂直两个方向的相关性，因此率失真性能要好一些。本文在拼接图像阵列下使用2D空间预测结构来压缩一幅由立体元图像阵列转换成的拼接图像阵列，与一维伪视频流方法和其它几种2D空间预测结构做比较，本文的方法要好一些。对组合成像技术采集到的视频序列的压缩，同样因对待图像的角度不同而产生了不同的方法。当把组合成像技术采集到的一幅静止图像视作具有周期性重复纹理的2D图像时，可是使用通用的2D视频压缩方案，如：MPEG-2/H.264/HEVC。当把它视作图像阵列时，可以使用多视点的方法压缩，图像阵列中每幅图像视作一个视点。本文把由多个立体元图像阵列组成的视频序列转换成由多个拼接图像阵列组成的视频序列，之后把拼接图像阵列中每一幅图像视作一个视点，使用多视点方法压缩。与直接使用HEVC压缩立体元图像阵列组成的视频序列方法做比较，本文的方法要好一些。更多还原

【Abstract】 Integral Imaging is an autostereoscopic3D display technology with the characteristicsof full color、continuous viewpoints，which is able to reproduce realistic3D scenes. Due to alarge amount of image data collected, Efficient compression to facilitate storage andtransmission is necessary. There are there common representation methods for a still imagecaptured by Integral Imaging: the direct captured element image array, the sub-image array,the ray-space image array. The sum-image array and ray-space image array can be obtainedfrom the element image array. This article describes a new type of representation, here called"spliced image array," and also obtained from the element image array. The beam overlapcondition of adjacent element images or adjacent sub-images from lens array and acquisitionpanel is analyzed. The beam overlap condition of adjacent images is one of the reasonswhich affect the similarity between adjacent images. The result shows that beam overlapratio of adjacent element images is positively correlated with the depth of3D object, so thesimilarity between adjacent element images may be increased. And beam overlap ratio ofadjacent sum-images is inverse correlated with the depth of3D object, so the similaritybetween adjacent sub-images may be reduced.When treating a still image captured by Integral Imaging in different ways, Thecompressing methods of the image may be different. The still image can be treated as awhole, one ordinary2D image with periodic repetitive texture. Or it can be treated as animage array(element image array or sub-image array or ray-space image array or splicedimage array). To compress an image array, there are several main methods: using transformaccording to the characteristics of image array; scanning the image array intoone-dimensional (1D) pseudo-video stream with scanning curve; using two-dimensional (2D)spatial prediction structure; compressed sensing and other methods. Scanning the imagearray into1D pseudo-video stream with scanning curve is equivalent to1D spatial predictionstructure, which only use the correlation of the image array in one direction(horizontal orvertical), therefore, the compression effect is not very good.2D spatial prediction structureuse the correlation of the image array in two direction(horizontal and vertical) at the sametime, therefore, the compression effect is better. This article uses2D spatial predictionstructure on spliced image array, compared with the1D pseudo video streaming method andseveral other2D spatial structure prediction methods, and the proposed method is better atleast for the test images.The compressing methods of one Integral Video also may be different when treating one still image captured by Integral Imaging in different ways. When one still image istreated as a whole, one ordinary2D image with periodic repetitive texture, a common2Dvideo compression schemes, such as: MPEG-2/H.264/HEVC can be used. When one stillimage is treated as an image array (element image array or sub-image array or ray-spaceimage array or spliced image array), the multi-view video method compression methods canbe used, and each image in image array is treated as one image in one viewpoint. This articletransform the Integral Video composed of element image array into the form composed ofspliced image array, then compress the Integral Video with the method for multi-view video，and one spliced image is treated as one image in one viewpoint. Compared with the directuse of HEVC compression method on Integral Video composed of element image array, theproposed method is better at least for the test images.更多还原