【作者】 许志涛；

【导师】 龙科慧； 刘金国；

【作者基本信息】 中国科学院研究生院(长春光学精密机械与物理研究所) ， 机械电子工程， 2014， 博士

【摘要】 随着空间遥感技术的发展，光学遥感图像分辨率越来越高，单位时间内获得的图像数据量越来越大。然而，空间遥感图像的传输和存储技术发展相对迟缓。因此，获得图像后，有必要对图像进行压缩编码处理。对于一幅图像而言，我们通常只关注其中的一部分区域或目标，即感兴趣区域，而其他区域称为背景区域。所以，在对图像进行压缩处理时，可以采用感兴趣区域压缩算法：对感兴趣区域进行无损压缩或低压缩比压缩，而对背景区域采用大压缩比压缩，从而既降低了图像传输对带宽的要求，又减少了感兴趣区域细节信息的丢失。论文主要研究一种基于CCSDS的空间遥感图像感兴趣区域压缩算法，并尝试采用基于视觉注意机制的itti模型来检测图像的感兴趣区域。论文以海洋监视卫星图像为研究对象，尝试采用itti模型来检测图像内的舰船目标等感兴趣区域。首先，研究了itti模型的算法处理过程：分析图像的多种特征，并将其融合生成特征显著图；然后采用胜者为王和返回抑制机制提取出视觉注意点；最后以该点为圆心，设置固定值为半径，划定圆形区域为显著区域。本文将视觉注意点的提取转移过程建立为电容阵列充电模型，并在算法中引入了离散矩变换，增强了图像纹理特征响应；由视觉注意点提取显著目标时，本文采用了阈值分割算法。实验结果表明，改进算法所提取的显著区域形状大小基本与目标一致，且显著区域包含背景少。与itti模型相比，改进算法更适合应用于海洋监视卫星图像舰船目标检测提取。本文探讨了SPIHT、JPEG2000以及CCSDS等图像压缩算法，并重点研究了CCSDS压缩标准。CCSDS将图像分为若干个不同的段，段与段之间独立编码，每段的纹理复杂度不同，所包含的信息量不同。本文采用梯度来衡量图像的纹理复杂度，并据此提出了一种基于梯度的压缩码流控制算法，纹理越复杂的段，所分配的码流容量越大，纹理越简单的段，所分配的码流容量越小。实验结果表明，采用该码流控制算法以后，恢复图像的信噪比有所改进。本文根据CCSDS的压缩特点，提出了一种新的感兴趣区域压缩算法，将感兴趣区域和背景区域进行分割，分别作为两幅独立的图像进行压缩。在压缩前，首先将感兴趣区域掩膜编码，然后将码流按一定比例分配给感兴趣区域和背景区域。之后引入基于梯度的码流分配算法，依次对感兴趣区域和背景区域编码，从而实现基于CCSDS的感兴趣区域图像压缩。实验结果表明，该算法能够提高图像感兴趣区域的恢复效果。更多还原

【Abstract】 With the development of space remote sensing technology, the resolution ofoptical remote sensing image is more and more high, and the data obtained in a shortperiod is increasing. However, the backward technology of transmission and storagefor remote sensing can’t meet the increasing image data. So, the images should becompressed before being transmitted. Usually, people only focus on a small part ofan image, which is called region of interest, while the other parts known asbackground region. When compressing an image, lossless compression or losscompression with a low ratio for the region of interest can be adopted, and thebackground region can be compressed using loss compression with a high ratio. As aresult, it not only reduces the requirements of the image transmission bandwidth, andalso reduces the loss of detail information of region of interest. The paper mainlystudies on ROI compression algorithm based on CCSDS, and tries to detect theregions of interest of remote sensing images with itti’s model, which is one of themodels based on visual attention mechanism.Itti′s model is applied to detect the ship targets, which is considered as theregion of interest of ocean surveillance satellite images. It illustrates the algorithmprocess of Itti′s model: firstly, the saliency map is obtained with the fusion of remote sensing image features, such as colors, intensity, orientations and so on; Secondly,the focus of attention is extracted using the mechanism of winner-take-all andinhibition of return; finally, setting the focus of attention as the center, a circularsalient region with a fixed radius is obtained. The paper introduces a capacitor arraycharging model to describe the extracting and transferring process of the focus ofattention, and also introduces the discrete moment transform to enhance the responseof image texture features. Then, the threshold segmentation method is chosen toextract the salient region with the focus of attention. it is verified that both the shapeand size of the salient region are consistent well with the ship targets; thebackground contained in the salient region is also reduced significantly. Moreover,the improved algorithm has a good real-time performance. It comes to the conclusionthat compared with Itti′s model, the improved algorithm is more effective andsuitable for the extraction of ship targets detection of ocean satellite images.This paper introduces SPIHT, JPEG2000, CCSDS and so on. CCSDS dividesthe image into several segments, and each segment is coded independently. Differentsegments contain different information, equaling to the image texture complexity,which is measured by gradient in the paper. According to the value of the gradient,the paper allocates the rate in the way that the bigger the value of a segment gradientis, the more rate of the segment. The experiment shows that, the rate-allocatingalgorithm is beneficial to optimize the rate distortion performance of CCSDS.According to the characteristics of CCSDS, this paper presents a new ROIcompression algorithm. After segmented, the region of interest and background ofthe image are compressed independently in the algorithm, which is implementedfollowing the steps: firstly, the ROI mask is coded; secondly, the rate is allocatedinto the regions of interest and background based on the value of each region; thirdly,the regions of interest and background are coded. The experiment shows that, thealgorithm can improve the recovery effect of the regions of interest.更多还原