基于自适应稀疏算子的图像乘性噪声移除方法研究

项目名称： 基于自适应稀疏算子的图像乘性噪声移除方法研究

项目编号： No.11526118

项目类型： 专项基金项目

立项/批准年度： 2016

项目学科： 数理科学和化学

项目作者： 周伟峰

作者单位： 青岛科技大学

项目金额： 2.5万元

中文摘要： 乘性噪声大量存在于医学成像、雷达成像和气象成像等，给相应的医学诊断、目标勘测和后续图像分析等带来困难。近年来，基于先验知识的字典学习型方法以其优越的图像复原质量的优势也被引入到乘性噪声移除问题。然而，此类方法模型的不适定性和与图像块对应的庞大学习基导致计算效率低下，而传统的固定基的稀疏算子(如传统DCT小波)自适应能力较弱使得成像效果参差不齐。鉴此，本项目拟通过构建基于先验退化图像的快速低维自适应稀疏算子来建立一类乘性噪声移除模型和快速算法，以便改进传统字典学习方法的计算效率低下和固定小波基的弱自适应性。本项目拟设计的稀疏算子将分别沿不同方向进行低维自适应扩散分解和各项同性重构，在保证图像质量的同时还提高了图像复原效率而且将易于推广到高维问题。

中文关键词： 稀疏表示；小波紧框架；乘性噪声；；

英文摘要： Multiplicative noise usually appears in medical imaging,radar and laser imaging, which effects badly the medical diagnosis,feature extraction and image analysis. Recent years, the dictionary learning methods based on prior knowledge have been introduced into multiplicative noise removal problems as a result of the superior image restoration quality. However,the learning bases based on large quantity of the prior images result in poor efficiency and the adaptive ability of the traditional fixed sparse operators(tranditional DCT wavelet) is weak. So this programme will design the adaptive sparse operator based on the prior degenerated images polluted by multiplicative noise and then investigate multiplicative noise removal model and fast algorithms based on the designed adaptive sparse operator in order to improve the poor efficiency of the learning based methods and the week adaptive ability of the fixed wavelet bases. The adaptive sparse operator designed will decompose along different directions and carry out isotropy reconstruction so the image restoration quality will be guaranteed and the efficiency will be improved simultaneously. Moreover, it will be easy to generalize this method to problems with high dimensions.

英文关键词： sparse representation；wavelet tight frame；multiplicative noise；；