| 基于自适应小波偏微分方程的蝗虫切片图像去噪 |
| |
| 引用本文： | 李丽,张楠楠,梅树立,李晓飞.基于自适应小波偏微分方程的蝗虫切片图像去噪[J].农业工程学报,2015,31(20):172-177. |
| |
| 作者姓名： | 李丽 张楠楠 梅树立 李晓飞 |
| |
| 作者单位： | 中国农业大学信息与电气工程学院，北京 100083,中国农业大学信息与电气工程学院，北京 100083,中国农业大学信息与电气工程学院，北京 100083,中国农业大学信息与电气工程学院，北京 100083 |
| |
| 基金项目： | 国家自然科学基金（41171337） |
| |
| 摘 要： | 蝗虫显微切片图像在获取的过程中不可避免地会受到噪声污染,其纹理、边缘与噪声又都属于高频分量,单独使用小波变换或偏微分方程(partial differential equation,PDE)扩散的方法都不能在有效去噪的同时保持边缘、纹理等。针对这一问题,提出了基于自适应小波PDE的去噪算法。首先对蝗虫切片含噪图像进行sym5小波软阈值去噪,分解层数根据去噪后图像的PSNR(peak signal to noise ratio)值自适应地选择,阈值门限使用Birge-Massart处罚算法获取。然后在此去噪的基础上进行Perona-Malik(PM)模型去噪,迭代次数根据去噪后图像的PSNR值自适应地选择,梯度阈值根据图像自身的2范数获取。为了验证所提出算法的去噪性能,进行了与常用去噪算法的对比试验。试验结果表明:视觉上,采用本文算法去噪后的图像噪声点较少且边缘、纹理清晰;客观上,采用该文算法去噪后的图像PSNR值比使用维纳滤波高出2 d B左右,比使用中值滤波高出3 d B左右,比使用小波阈值去噪高出2 d B左右,比使用PM模型去噪高出1 d B左右,并且在结构相似性(structural similarity image measurement,SSIM)上采用该文算法去噪后的图像与原始图像的相似度最高。因此,将自适应小波PDE的算法应用于蝗虫切片去噪是可行的、有效的,为其后续处理提供了技术支持。
|
| 关 键 词： | 切片 图像 小波 蝗虫切片 图像去噪 PM模型 结构相似性 |
| 收稿时间： | 2015/7/28 0:00:00 |
| 修稿时间： | 2015/9/20 0:00:00 |
Image de-noising of locust sections based on adaptive wavelet and partial differential equation method |
| |
| Li Li,Zhang Nannan,Mei Shuli and Li Xiaofei.Image de-noising of locust sections based on adaptive wavelet and partial differential equation method[J].Transactions of the Chinese Society of Agricultural Engineering,2015,31(20):172-177. |
| |
| Authors: | Li Li Zhang Nannan Mei Shuli and Li Xiaofei |
| |
| Institution: | College of Information and Electrical Engineering, China Agricultural University, Beijing 100083, China,College of Information and Electrical Engineering, China Agricultural University, Beijing 100083, China,College of Information and Electrical Engineering, China Agricultural University, Beijing 100083, China and College of Information and Electrical Engineering, China Agricultural University, Beijing 100083, China |
| |
| Abstract: | Abstract: Noise pollution on locust micro-section images is always unavoidable during the acquisition of the images. However, few researches have been devoted to the de-noise processing of locust section images. The locust section image is generally characterized by rich textures, smooth regions and well-defined edges. Since the textures, the edges and noises of the images are high-frequency components, wavelet transformation can't successfully get rid of noise on the images effectively without destroying the edge features, i.e., it might cause the pseudo-Gibbs' effect and edge blurring. Since the gradient value of the textures is small while the gradient value of the edges and noises is large, partial differential equation (PDE) diffusion can't successfully get rid of noise on the images effectively without destroying the texture, i.e., it tends to lose the original textural details. Therefore, we proposed a new algorithm for the de-noise of locust section image, which was called adaptive wavelet PDE method. It possessed all the advantages of wavelet decomposition and anisotropic diffusion. It could remove noises successfully with the textural details preserved and the edges clear. The procedure of the proposed algorithm included 2 steps as follows. First, we de-noised the images using the sym5 wavelet soft-threshold algorithm, in which the wavelet decomposition level was adaptively selected according to the PSNR (peak signal to noise ratio) value of the de-noise images and the soft-threshold was obtained by the Birge-Massart penalty algorithm. Further de-noising was done with the Perona Malik (PM) model, in which the iterations were adaptively selected according to the PSNR value of the de-noise images, and the gradient threshold according to the 2-norm of the image grey value. After the implementation of the adaptive wavelet PDE algorithm, a 3-step simulation test was made to evaluate the effectiveness of the proposed algorithm using MATLAB 8.2. In order to determine the optimal wavelet decomposition level for the image, we compared the image de-noising results on different wavelet decomposition levels. The experiments showed that wavelet decomposition level should be 2 while using the wavelet soft-threshold for the de-noise image. Then, to determine the optimal iterations for the PM model, the de-noise results in different iterations were compared with each other. The experiments showed that the iterations between 5 and 10 (inclusively) were appropriate while using the PM model for the de-noise image. Finally, the proposed algorithm had some comparison with the conventional de-noise algorithms. The de-noised image obtained by the proposed algorithm was less residual noise and clearer textures than other algorithms visually. We used 2 common de-noise evaluation criteria of image, i.e. PSNR and structural similarity image measurement (SSIM), which measured the degree of image distortion and similarity between the processed and the original image. The PSNR value of de-noise image obtained by the proposed algorithm was 28.7474 dB, which was higher than using the Wiener filtering, the median filtering, the wavelet threshold de-nosing and the PM Model de-nosing, by 2, 3, 2 and 1 dB, respectively. It was higher than the PSNR value of the noisy image by 4.6 dB. The SSIM value of de-noise image using the proposed algorithm was 0.8258, which was the largest among the above-mentioned algorithms and this indicated the de-noise image using the proposed algorithm was closer to the original image in the brightness, contrast and structure aspects. In conclusion, the proposed algorithm is feasible and effective for de-noising locust section image. It will provide technical support to the subsequent processing of the image, which will bring convenience to better understand the structure of locust cells and nerves and hence be helpful to reduce pollution resulting from the abuse of chemical pesticides ultimately. |
| |
| Keywords: | section image wavelet locust section image de-noising Perona-Malik model structural similarity image measurement (SSIM) |
| 本文献已被 CNKI 等数据库收录！ |
| 点击此处可从《农业工程学报》浏览原始摘要信息 |
| 点击此处可从《农业工程学报》下载免费的PDF全文 |
|
