基于云数据的高效图像编码方法

发布时间：2018-06-24 08:28

本文选题：大数据 + 云编码　；参考：《计算机学报》2017年11期

【摘要】：随着数字图像采集设备的普及,加之互联网技术的高速发展,海量的图像在上传并保存到互联网中.传统的图像编码方法(如JPEG和JPEG2000),可以在一定程度上压缩图像数据,但其独立压缩单幅图像的方式导致其压缩效率远远不能满足当今大数据的需求.为了更高效地压缩互联网图像数据,作者充分利用"大数据"这一挑战,提出了一种基于云数据的图像编码方法.针对上传到云数据库中的图像,算法采用"上传一幅、压缩一幅"的处理模式,有效地适应迅速变化的大数据环境.它充分利用了当前图像和云数据中其它图像之间的相关性,更加有效地对图像进行压缩编码.在该方法中,作者有3个创新性技术设计.首先,基于高效的图像检索技术,设计了一个方案对"大数据"进行筛选,仅保留与当前图像高度匹配的少部分图像数据.其次,利用这些经过选择得到的少量图像数据,进行图像间的预测编码.这一环节对于整个系统的性能非常重要.作者采用基于块匹配的预测编码和率失真优化技术,在保证图像重建质量的前提下将码率大大减少.还有,该文设计了一套对检索得到图像的预处理方案,将其变换成更为接近于当前图像的参考图像,从而提高预测准确度.作者采用投影变换技术将检索得到的图像与当前图像对齐,并通过光照补偿使得两幅图像中的匹配像素点的值尽可能接近.该文为大数据下的图像压缩提供了一个崭新的框架,可以大大节省如网络云盘中图像所占用的存储空间.相比于传统图像编码方法甚至是更为高效的HEVC帧内编码方案,该图像压缩系统都能取得明显的性能提升.实验表明:该方法对图像的压缩效率,比JPEG平均提升78.5%,比HEVC帧内编码提升67.2%.甚至相比于文献中最新的基于云的图像编码方法,该文的算法在同码率下的主观质量和客观质量都表现出明显优势.
[Abstract]:With the popularity of digital image acquisition equipment and the rapid development of Internet technology, massive images are uploaded and stored on the Internet. Traditional image coding methods (such as JPEGs and JPEG2000) can compress image data to a certain extent, but their compression efficiency is far from satisfying the demand of big data. In order to compress the Internet image data more efficiently, the author takes full advantage of the challenge of "big data" and proposes an image coding method based on cloud data. For the images uploaded to the cloud database, the algorithm adopts the processing mode of "upload one image, compress one image", and adapt effectively to the rapidly changing big data environment. It makes full use of the correlation between the current image and other images in cloud data, so it can compress and encode the image more effectively. In this method, the author has three innovative technical designs. Firstly, based on the efficient image retrieval technology, a scheme is designed to filter "big data", and only a few image data matching the current image height are retained. Secondly, using these selected small amount of image data, the prediction coding between images is carried out. This link is very important to the performance of the whole system. The prediction coding based on block matching and rate-distortion optimization techniques are used to reduce the bit rate greatly while ensuring the quality of image reconstruction. In addition, this paper designs a preprocessing scheme for image retrieval, which is transformed into a reference image closer to the current image, so as to improve the prediction accuracy. The author uses projection transform technique to align the retrieved image with the current image, and makes the matching pixel value of the two images as close as possible through illumination compensation. This paper provides a new frame for image compression under big data, which can save the storage space such as the image in the cloud disk. Compared with the traditional image coding method and even the more efficient HEVC intra coding scheme, this image compression system can achieve a significant performance improvement. Experimental results show that the efficiency of the proposed method is 78.5% higher than that of JPEG and 67.2% higher than that of HEVC intra coding. Compared with the latest cloud-based image coding methods in the literature, the proposed algorithm has obvious advantages in both subjective and objective quality at the same bit-rate.
【作者单位】：北京大学信息科学技术学院;南洋理工大学博云搜索实验室;
【基金】：国家“九七三”重点基础研究发展规划项目基金(2015CB351800) 国家自然科学基金优秀青年科学基金(61322106)、国家自然科学基金(61370114)资助~~
【分类号】：TN919.81

【相似文献】