一种图像感兴趣区域提取方法研究  

Investigation on Extraction of Interests Region in Image

作　　者：王诚[1] 范向阳[1] 

机构地区：[1]南京邮电大学通信与信息工程学院,江苏南京210003

出　　处：《计算机技术与发展》2017年第8期30-36,共7页Computer Technology and Development

基　　金：国家自然科学基金资助项目(61071167)

摘　　要：关于图像感兴趣区域(ROI)提取,改进的Stentiford视觉模型方法与传统的Stentiford视觉模型方法以及其他视觉模型方法相比,具有提取的图像区域清晰、边缘明显、效率高等优点,但在图像背景较为复杂时,会提取到目标区域以外的区域。鉴于实际研究中对准确度的要求,需要从已提取区域中挑选出目标区域。为此,在所涉及的图像处理过程中,提出了多种图像新特征的提取方法,并引入数据挖掘领域中的经典FP-Growth算法,在改进的Stentiford视觉模型方法对训练集图像处理后,提取图像显著熵、显著熵密度等众多图像特征,并应用FP-Growth算法挖掘图像特征和目标区域的关联规则,同时将获取到的规则应用于测试集的大量实验验证之中。实验结果表明,采用了所提出的方法后,提取到的图像区域准确度有显著提高,表明该方法是可行有效的。Concerning extraction of Region Of Interest （ROI） for image, the enhanced Stentiford visual modeling method has the advanta- ges of clear image regions, obvious edges, and high efficiency compared with traditional Stentiford or other visual modeling methods. However, when the image background is relatively complex, regions outside the target would be extracted. Because of requirements on ac- curacy in actual research, it is necessary to pick out target regions from extracted regions. For this reason, during image processing, an ex- traction method with multiple new image features has been proposed and the classic FP-Growth algorithm in data mining has been intro- duced. After the training set have been processed by the enhanced Stentiford visual modeling method, image features such as significant entropy and significant entropy density are extracted and the FP-Growth algorithm is employed to find the association roles between im- age features and target regions. The obtained roles have been applied to a lot of experimental for verifications. The experimental results show that after the enhanced method, the accuracy of the extracted image regions have been improved significantly which indicates its fea- sibility and effectiveness.

关 键 词：图像感兴趣区域 Stentiford视觉模型 FP—Growth算法 图像特征 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]