【作者】 王然；

【导师】 胡瑞敏；

【作者基本信息】 武汉大学 ， 通信与信息系统， 2019， 硕士

【摘要】 近年来,随着遥感技术的不断发展,高分辨率对地观测卫星已成为国家重要战略资源之一。由于卫星所采集的图像不同于自然场景图像,卫星图像具有幅宽大、精度高、背景复杂等特点,在极端的星上资源环境与有限的星地传输带宽下,未经处理的卫星图像数据会给星上存储器和处理器带来极大的挑战,因此,需要在星上引入图像压缩系统。为了满足星地传输的需求,卫星图像通常需要采用比自然场景图像更大的压缩率。传统图像压缩方法对所有最小编码单元采用相同的压缩方案,难以保障低码率下图像的质量。同时,处理大幅宽卫星图像使得压缩过程所需时间增加,难以满足卫星图像低延时传输的需求。卫星图像的应用不在于观赏,而是对图像内容的理解与分析。虽然卫星图像的幅宽大、背景较为复杂,但是图像中的某些区域仍具有提取并且分析的价值,这些区域则被称为卫星图像的显著区域。对卫星图像进行显著性检测,能够使人们最大限度地忽略无关信息,提取源图像中的重要元素。图像压缩中的质量和时间优化,通常依靠对压缩过程中的码率控制过程的改进。在码率控制过程中,对显著区域分配更多的码率,对非显著区域减少冗余压缩操作,能够有效改善卫星图像的压缩质量,并减少码率控制的时间复杂度。目前图像的显著性检测方法面向自然场景图像,背景简单,显著区域单一且集中,不适用于卫星图像。同时各图像压缩方法对其最小编码单元均采取相同的码率控制过程,普遍没有考虑图像的内容的不同重要性,未对码率做出按需分配的操作,因此图像压缩的整体效率较低。针对传统码率控制方法的不足,本文以JPEG2000图像压缩标准为基础,提供了一种面向卫星图像压缩的码率控制方法,利用卫星图像的显著性检测对码率控制过程的图像质量、压缩时间进行改进优化。在本文提出的方法中,首先利用卫星图像的特点,在主流的图像显著性检测方法中筛选最适用于卫星图像的方法,并对其进行改进处理;之后,利用所得到的显著图,对码率控制过程做出压缩质量、时间两方面的优化:压缩质量优化作用于码率控制过程中的量化阶段,利用显著性检测进行量化步长的加权分配,以达到按图像内容重要性按需分配码率的目的;压缩时间优化作用于熵编码阶段,利用显著性检测进行每个位点的编码原语的分配,避免三次编码通道扫描过程,跳过无显著性的位点,减少码率控制过程的时间冗余。本文通过码率控制准确度、主客观质量、算法耗时这三个指标来检测所提出方法的有效性。实验结果证明,所提出方法具有良好的码率控制准确度,并且低码率下,显著区域的图像峰值信噪比较原始JPEG2000约有1.82dB的提升,算法耗时约有23%的节省。更多还原

【Abstract】 With the rapid development of remote sensing technology,high-resolution earth observation satellites have become one of the important strategic resources of the country.Since the image acquired by the satellite is different from that of natural scenes,the satellite image has the characteristics of large width,high precision and complex background.Under the extreme satellite resource environment and limited satelliteearth transmission bandwidth,the unprocessed satellite image data will cause great challenges to on-board memories and processors,so an image compression system needs to be introduced onto the satellite.In order to meet the demand of satellite-earth transmission,satellite images usually need to adopt a larger compression rate than natural scene images.The traditional image compression method adopts the same compression scheme for all minimum encoding units,which is difficult to guarantee the image quality under low bit rate.At the same time,processing large-width satellite images increases the time required for the compression process,which is difficult to meet the demand of low latency transmission of satellite images.The application of satellite image is not about viewing,but about the understanding and analysis of image contents.Although the satellite image has a large width and a complex background,some regions of the image still are valuable for extraction and analysis,which are called salient regions of the satellite image.And the saliency detection on satellite images enables people to minimize irrelevant information and extract important elements from the source image.The quality and time optimization in image compression usually depends on the improvement of the bit-rate control process.Allocating more bit rate on salient region while reducing redundant compression operations for non-salient region in the process of bit-rate control,can effectively improve the compression quality of satellite image and reduce the time complexity of bit-rate control.At present,the saliency detection method of image is oriented to natural scene images,in which the background is simple and the salient region is single and centralized.Therefore,it is not suitable for satellite images.At the same time,each image compression method adopts the same bit-rate control process for its minimum encoding unit,which generally does not consider the different importance of the content of images,and does not conduct the on-demand allocation to the bit rate,so the overall efficiency of image compression is relatively low.Aiming at the shortcomings of the traditional rate control method,this paper provides a rate control method for satellite image compression based on the JPEG2000 image compression standard.By the saliency detection of satellite image,this method can improve and optimize the quality and compression time of the rate control process.The method,according to the characteristics of satellite images,firstly screens the most suitable method for satellite images in the mainstream image saliency detection ones,and then processes this suitable method;after that,the obtained saliency image is used to the rate control process for optimizing both compression quality and time—compression quality optimization is applied to the quantization phase in the rate control process,in which saliency detection is used to perform weighted allocation of quantization step to achieve the purpose of the on-demand distribution of rate according to image content importance;the compression time optimization acts on the entropy coding stage,in which the saliency detection is used to allocate the coding primitives of each pixel,avoiding the scanning process of the three coding passes,skipping the non-salient pixels,and reducing the rate control time redundancy of the process.In this paper,the effectiveness of the proposed method is tested by three factors: rate control accuracy,subjective and objective quality,and algorithm time consumption.The experimental results show that the proposed method has good rate control accuracy,and the PSNR of the salient regions in the image is about 1.82 dB higher than that of the original JPEG2000 at low bit rate.The algorithm takes about 23% less time than that of existing methods.更多还原