计算机应用 ›› 2011, Vol. 31 ›› Issue (03): 745-748.DOI: 10.3724/SP.J.1087.2011.00745

• 图形图像技术 • 上一篇    下一篇

基于小数目标尺度的图像混合滤波算法

钱晓亮,郭雷,余博   

  1. 西北工业大学 自动化学院,西安710129
  • 收稿日期:2010-09-01 修回日期:2010-10-19 发布日期:2011-03-03 出版日期:2011-03-01
  • 通讯作者: 钱晓亮
  • 作者简介:钱晓亮(1982-),男,河南孟州人,博士研究生,主要研究方向:特征提取、智能算法;郭雷(1956-),男,山东海阳人,教授,博士生导师,主要研究方向:神经计算、图像和视频处理;余博(1980-),男,陕西西安人,博士研究生,主要研究方向:目标识别与跟踪。
  • 基金资助:
    航空科学基金资助项目(20080153002)

Hybrid image filter based on decimal object scale

QIAN Xiao-liang,GUO Lei,YU Bo   

  1. School of Automation, Northwestern Polytechnical University, Xi'an Shaanxi 710129, China
  • Received:2010-09-01 Revised:2010-10-19 Online:2011-03-03 Published:2011-03-01
  • Contact: QIAN Xiao-liang
  • Supported by:
    the Aeronautical Science Foundation of China under Grant

摘要: 为了在有效去除可见光图像噪声的同时最大限度地保持图像的边缘、纹理等细节,将已有的目标尺度改进为小数目标尺度以便更精确地反映局部目标结构的大小,提出了基于小数目标尺度的自适应高斯滤波和基于小数目标尺度的自适应中值滤波的混合滤波算法。前者通过小数目标尺度来自适应地控制高斯核的尺度和滤波的模板大小,后者利用小数目标尺度自适应地筛选出脉冲噪声点并进行中值滤波,并弥补前者在抑制脉冲噪声方面的不足。理论分析和仿真实验结果均表明,所提出的算法不仅可以去除各种类型的点状噪声,而且在图像细节的保护和信噪比方面优于其他几类传统算法。

关键词: 小数目标尺度, 自适应高斯滤波, 自适应中值滤波, 点状噪声

Abstract: To remove the noise of optical images while preserving its fine details, the extant object scale was upgraded to the decimal object scale for reflecting the size of local object structure more accurately, and a hybrid image filter which contains two parts was proposed. The first part was an adaptive Gaussian filter based on decimal object scale, the scale of the Gaussian kernel and the mask size of filtering were controlled adaptively by the decimal object scale. The second part was an adaptive median filter based on decimal object scale, and the impulse noise points which were selected adaptively by the decimal object scale were filtered. The weakness of the first part in suppressing the impulse noise was remedied by the second part. Both theory analysis and simulation results show that the presented method can suppress various point-like noise and it is superior to several traditional methods in preserving the fine details and signal to noise ratio.

Key words: decimal object scale, adaptive Gaussian filter, adaptive median filtering, point-like noise

中图分类号: