通信学报

• 论文II • 上一篇    下一篇

图像分块压缩感知中的自适应测量率设定方法

李 然,干宗良,崔子冠,武明虎,朱秀昌   

  1. 南京邮电大学 江苏省图像处理与图像通信重点实验室,江苏 南京 210003
  • 出版日期:2014-07-25 发布日期:2014-07-15
  • 基金资助:
    国家自然科学基金资助项目(61071091, 61271240);江苏省研究生创新计划基金资助项目(CXZZ12_0466, CXZZ11_0390);江苏省高校自然科学研究基金资助项目(12KJB510019);南京邮电大学校科研基金资助项目(NY212015);湖北省教育厅科研重点基金资助项目(D20121408)

Adaptive measurement rate setting method in block compressed sensing of images

  • Online:2014-07-25 Published:2014-07-15

摘要: 传统的图像分块压缩感知(BCS, block compressed sensing)以相同的测量率对各块进行测量,但由于图像的空间特性不同,在重构图像时出现了块效应。通过自适应为各块设定不同的测量率,该问题可得到有效的解决。然而,已有的自适应测量率设定法需要在采集端获得原始数字图像,这在实际的压缩成像(CI, compressive imaging)设备中无法实现。为了克服这一缺陷,提出了一种更易于通过硬件实现的自适应测量率设定法。该方法利用在采集端可获得的CS测量值直接在测量域中估计各图像块的样本方差,再根据各块样本方差自适应地为每块设定测量率并实现码率控制。仿真实验结果表明,该方案重构图像的质量优于非自适应方案,但由于测量域估计块样本方差存在偏差,使其与直接利用块样本方差真实值的自适应方案相比,仍具有一定差距。

Abstract: Traditional block compressed sensing (BCS) of images uses the same measurement rate to measure each block, but some blocking artifacts appear in the reconstructed image on accounting of varying spatial characteristics in an image. This problem can be effectively solved by adaptively setting different measurement rate for every block. However, these existing methods require original digital image at the collector, which cannot be realized by using practical compressive imaging (CI) devices. In order to overcome this shortage, an adaptive measurement rate setting method is proposed and it can be easily achieved though hardware equipments. This method uses the CS measurements acquired at the collector to estimate the sample variance of each block directly, and then adaptively sets measurement rate of each block in terms of their sample variances and realize rate control. Experimental results show that proposed method can obtain a better quality of reconstructed image than non-adaptive scheme, but there is a gap between proposed method and the adaptive scheme using the true block sample variance since the sample variance estimated in the measurement domain has some deviations.

No Suggested Reading articles found!