| 引用本文: | 翟永平,周东翔,刘云辉.基于颜色及梯度统计特征的结核杆菌目标识别.[J].国防科技大学学报,2012,34(5):146-152.[点击复制] |
| ZHAI Yongping,ZHOU Dongxiang,LIU Yunhui.Recognition of mycobacterium tuberculosis in microscopic images based on color and gradient feature[J].Journal of National University of Defense Technology,2012,34(5):146-152[点击复制] |
|
| |
|
|
| 本文已被：浏览 8635次 下载 6397次 |
| 基于颜色及梯度统计特征的结核杆菌目标识别 |
| 翟永平1, 周东翔1, 刘云辉2 |
|
(1.国防科技大学 电子科学与工程学院,湖南 长沙 410073;2.香港中文大学 机械与自动化工程系,中国 香港)
|
| 摘要: |
| 提出一种基于颜色及梯度统计特征的结核杆菌目标识别算法。首先基于HSV颜色空间进行图像预分割，然后在CIE L*a*b*颜色空间进行自适应分割以提取目标精细几何形状。为了适应背景的复杂变化，基于色调一致性假设对疑似目标进行色调一致性检验并剔除大部分伪目标。为了将重叠粘连目标从伪目标中分离出来，提出两个梯度统计特征，然后结合目标的面积、周长、长宽比、圆形度、粗糙度等形态特征，组成7维特征向量送入贝叶斯分类器进行分类。实验结果表明，本文算法能适应标本、染色及图像背景等复杂变化，目标识别率可达91%。 |
| 关键词: 结核杆菌 显微图像 HSV颜色空间 贝叶斯分类器 形状描述子 梯度特征 |
| DOI： |
| 投稿日期：2012-03-05 |
| 基金项目:国家自然科学基金资助项目(60975023) |
|
| Recognition of mycobacterium tuberculosis in microscopic images based on color and gradient feature |
| ZHAI Yongping1, ZHOU Dongxiang1, LIU Yunhui2 |
|
(1.College of Electronic Science and Engineering, National University of Defense Technology, Changsha 410073, China;2.Department of Mechanics and Automation Engineering, The Chinese University of Hong Kong, Hong Kong, China)
|
| Abstract: |
| A color and gradient feature based image segmentation and recognition algorithm is proposed for Mycobacterium Tuberculosis (TB) objects in microscopic images. First, the input color image was pre-segmented based on the HSV color space using the threshold method according to the prior information. Then, the original image was transformed to the CIE L*a*b* color space, and the L component image was segmented using an adaptive threshold method to get finer segmentation result. In order to accommodate the complex variety of image background, all the suspected objects were verified according to the “hue consistence” assumption and the false objects were rejected. To identify the TB bacilli, the algorithm used five shape feature descriptors, including the area, the perimeter, the ratio of width to height, the compactness and the roughness, and two gradient feature descriptors, the gradient magnitude weighted average (GMWA) and the gradient magnitude variance average (GMVA) respectively, and makes the judgment through Bayes classifier. Experimental results show that the proposed algorithm can accommodate the complex variety of specimens and the image background, and a high recognition rate (91%) can been obtained. |
| Keywords: mycobacterium tuberculosis microscopic image HSV color space Bayes classifier shape feature descriptor gradient feature |
|
|
|
|
|
