基于联合特征PCANet的宫颈细胞图像分类识别方法研究

发布时间：2018-03-12 16:37

本文选题：图像去噪　切入点：图像增强　出处：《广西师范大学》2017年硕士论文　论文类型：学位论文

【摘要】：宫颈癌日益威胁着广大女性的健康,因而宫颈癌的早期筛查预防就显得非常必要,计算机辅助自动化诊断可以有效减少人工对宫颈细胞图像的判读的误差,并降低人工成本,使宫颈癌筛查技术可以快速推广,具有很好的社会价值和经济效益。本文对宫颈细胞图像分类识别方法的关键技术进行了研究,包括宫颈细胞图像去噪,增强,特征提取和分类识别。主要研究内容为:(1)采用基于块组的非局部自相似先验学习图像去噪(patch group based nonlocal self-similarity prior learning for image denoising,PGPD)方法用于宫颈细胞图像去噪处理。仿真实验表明本文所采用的去噪方法对宫颈细胞图像去噪的同时能够较好地保护宫颈细胞图像的结构信息,且在噪声增加时峰值信噪比(peak signal to noise ratio,PSNR)与结构相似性指数(structural similarity index,SSIM)降低的程度较小,因而具有较好的鲁棒性。(2)采用基于自适应S型函数的B直方图均衡方法对宫颈细胞图像进行增强处理,使得图像特征突出,有利于特征提取。(3)在PCANet的基础上构造联合特征PCANet将网络中间层提取的特征与最后一层输出的特征联合起来作为最终的特征输出,联合特征PCANet可以减少图像特征在逐层提取过程中的丢失,因而使最后提取的特征能更好地表征图像之间的差异。得到提取的特征后再利用SVM进行分类识别。仿真实验表明本文方法对宫颈细胞图像二分类识别准确率为95.71%,三分类识别准确率为85.40%,具有一定应用价值。(4)基于MATLAB GUI设计了宫颈细胞图像分类识别系统,包含训练和检测两个模块,实现联合特征PCANet网络和分类器的训练以及宫颈细胞图像的检测,功能实现完整,操作简洁,具有较好的应用价值。
[Abstract]:Cervical cancer is increasingly threatening the health of women, so the early screening and prevention of cervical cancer is very necessary. Computer aided automated diagnosis can effectively reduce the error of artificial interpretation of cervical cell images, and reduce the cost of labor. So that cervical cancer screening technology can be quickly popularized, with good social value and economic benefits. This paper studies the key technology of cervical cell image classification and recognition, including cervical cell image denoising, enhancement, Feature extraction and classification recognition. The main content of the study is: 1) using the non-local self-similar priori learning image denoising patch group nonlocal self-similarity prior learning for image DenoisingPGPDs method for cervical cell image denoising. The simulation results show that the proposed method can be used in cervical cell image denoising. The method used in this paper can protect the structure information of cervical cell image while removing noise from cervical cell image. The decrease of peak signal to noise PSNRs and structural similarity index (SSIMI) was smaller when noise increased. Therefore, the method of B-histogram equalization based on adaptive S-function is used to enhance the cervical cell image, which makes the image feature prominent. It is advantageous to construct a joint feature based on PCANet. PCANet combines the features extracted from the middle layer of the network with the features of the last layer as the final feature output. Joint feature PCANet can reduce the loss of image features in the process of layer by layer extraction. As a result, the extracted features can better represent the differences between the images. The extracted features are then classified and recognized by SVM. The simulation results show that the accuracy of the method is 95.71 for cervical cell images. The accuracy of three classification recognition is 85.40, which has certain application value. (based on MATLAB GUI, a cervical cell image classification and recognition system is designed. It includes two modules: training and detecting, which realizes the training of joint feature PCANet network and classifier and the detection of cervical cell image. The function is complete, the operation is simple, and it has good application value.
【学位授予单位】：广西师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R737.33;TP391.41

【参考文献】