Practical Optimization Algorithms for Large-Scale Image and Data Processing

大规模图像和数据处理的实用优化算法

• 批准号: 0811188
• 负责人: Yin Zhang
• 金额: $ 24.2万
• 依托单位: William Marsh Rice University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2008
• 资助国家: 美国
• 起止时间: 2008-09-15 至 2012-02-29
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0811188&HistoricalAwards=false
• 关键词: Practical; Optimization; Algorithms; Large; Scale

Optimization algorithms are at the core of solving many problems in image and data processing, and dedicated algorithms are often critical in real-world applications. The principal investigator (PI) will conduct algorithmic research in two important areas: image deconvolution and compressive sensing, to develop enabling algorithms that make relevant methodologies practical for large-scale, real applications. For image deconvolution, the PI aims to develop optimization algorithms for total-variation-based models that are faster than existing algorithms by at least one or more order of magnitude. Preliminary studies have shown that this ambitious goal is well within grasp. The new compressive sensing(CS) methodologies make it possible to significantly reduce the number of measurements needed for reconstructing compressible data. The PI proposes to develop algorithms for important real-world applications of CS, and study random Kronecker-product measurement matrices that can drastically reduce data reconstruction complexity.MRI (magnetic resonance imaging) is a widely used medical imaging modality that creates an image from scanned data. A typical abdominal scan may take around 90 minutes. Recent progress in a new methodology called compressive sensing (CS) makes it possible to reduce this time to 30 minutes by scanning only one third of data, while maintaining good image quality. However, such a possibility can be realized only when fast algorithms are available to do real-time processing on incomplete data. This project is to develop and analyze such fast algorithms. Another class of fast algorithms to be investigated is for improving the clarity of fuzzy images. With such fast algorithms, for example, satellite or medical images can be better analyzed in a more timely fashion. The results of this project will impact applications ranging from information technology to biotechnology.

优化算法是解决图像和数据处理中许多问题的核心，而专用算法在实际应用中通常至关重要。 首席研究员（PI）将在两个重要领域进行算法研究：图像反卷积和压缩感知，以开发使相关方法可用于大规模实际应用的支持算法。 对于图像反卷积，PI 的目标是为基于全变分的模型开发优化算法，该算法比现有算法快至少一个或多个数量级。 初步研究表明，这一雄心勃勃的目标是完全可以实现的。新的压缩感知（CS）方法可以显着减少重建可压缩数据所需的测量数量。 PI 提议为 CS 的重要实际应用开发算法，并研究随机克罗内克积测量矩阵，该矩阵可以大大降低数据重建的复杂性。MRI（磁共振成像）是一种广泛使用的医学成像模式，可从扫描图像中创建图像数据。 典型的腹部扫描可能需要大约 90 分钟。 压缩传感 (CS) 新方法的最新进展使得只需扫描三分之一的数据即可将时间缩短至 30 分钟，同时保持良好的图像质量。 然而，只有当快速算法能够对不完整数据进行实时处理时，这种可能性才能实现。该项目旨在开发和分析此类快速算法。 另一类需要研究的快速算法是提高模糊图像的清晰度。 例如，利用如此快速的算法，可以更及时地更好地分析卫星或医学图像。 该项目的结果将影响从信息技术到生物技术的各种应用。