| 引用本文: | 吴玲达,郝红星.一种图像去模糊正则化恢复算法参数确定方法.[J].国防科技大学学报,2012,34(4):79-84.[点击复制] |
| WU Lingda,HAO Hongxing.Research on the regularization parameter of image deblurring[J].Journal of National University of Defense Technology,2012,34(4):79-84[点击复制] |
|
| |
|
|
| 本文已被：浏览 9386次 下载 7636次 |
| 一种图像去模糊正则化恢复算法参数确定方法 |
| 吴玲达1,2, 郝红星1 |
|
(1.国防科技大学 信息系统与管理学院，湖南 长沙 410073;2.2.装备学院 复杂电子系统仿真重点实验室，北京 怀柔 101400)
|
| 摘要: |
| 为了解决模糊正则化恢复算法中参数确定问题，提出一种正则化参数确定方法，该方法根据降质图像特征计算正则化参数。分析了目前普遍应用的全变分正则化方法和该问题的改进拉格朗日迭代解法（SALSA），分析不同正则化参数对恢复效果的影响，提出的正则化参数确定方法与噪声和原图像梯度大小相关。对不同梯度和噪声图像的不同正则化参数恢复效果进行对比，得到提出的正则化参数确定方法能使恢复图像的改进信噪比处于最大值附近。从实验视觉效果得出，该参数确定方法能够抑制降质图像的噪声并能够尽量恢复原图像细节信息。 |
| 关键词: 正则化参数 去模糊 全变分正则化 改进拉格朗日迭代解法 |
| DOI： |
| 投稿日期：2012-01-15 |
| 基金项目:国家自然科学基金资助项目（61103081）；国防科技大学优秀博士生创新资助项目（B110503） |
|
| Research on the regularization parameter of image deblurring |
| WU Lingda1,2, HAO Hongxing1 |
|
(1.College of Information System and Management, National University of Defense Technology, Changsha 410073, China;2.2.Key Laboratory of Science and Technology for National Defense, College of Equipment, Beijing 101400, China)
|
| Abstract: |
| A method of determining the regularization parameter is proposed in order to solve the image deblurring problems. The main objective is to get the regularization parameter from the deblurring image. The method starts by considering the famous Total Variation methods and the Split Augmented Lagrangian Shrinkage Algorithm (SALSA), and then it deals with the effect of the regularization parameter on the result image. The proposed method proves that the parameter is determined by the noise and the gradient of the image. Results of the recovery of images were compared with different gradient and noise, and they help to reach a conclusion that the proposed method maximizes the improvement in Signal to Noise Ratio (ISNR). The method also removed noise and protected the details of the image in the processing of recovery as much as possible. |
| Keywords: regularization parameter image deblurring total variation augmented Lagrangian methods |
|
|
