Image-Analysis:STATGR5293

  • H9_470625
    了解作者
  • 3.7MB
    文件大小
  • zip
    文件格式
  • 0
    收藏次数
  • VIP专享
    资源类型
  • 0
    下载次数
  • 2022-06-08 03:49
    上传日期
图像分析:探索图像分析及其应用 人脸检测 使用基于颜色的k均值聚类检测图像中的人脸。 虹膜识别 John Daugman的虹膜检测算法和Li Ma的空间滤波器的实现,以检测和识别虹膜。 1.虹膜定位 使用Canny边缘检测和Hough变换在由瞳Kong中心确定的特定区域中获取瞳Kong和虹膜圆的确切参数。 返回虹膜和瞳Kong的坐标和半径信息。 2.虹膜归一化 使用Li Ma的论文中提供的方法将虹膜从笛卡尔坐标映射到极坐标。 返回尺寸为68 x 512的归一化图像。 3.图像增强 归一化的虹膜图像具有低对比度,并且可能由于光源的位置而导致亮度不均匀。 我们通过图像的每个32 x 32区域中的直方图均衡化来增强图像。 Li Ma的“关注区域”仅包含图像的48 x 512部​​分,因此我们仅返回此ROI的增强图像。 4.特征提取 Gabor过滤器已用于过滤图像,二维卷积已用于将原始图像与过滤
Image-Analysis-master.zip
  • Image-Analysis-master
  • Brain State Classification
  • MATLAB
  • Figure 30.4.png
    320.2KB
  • SPM.mat
    1.3MB
  • Figure 30.10.png
    109.9KB
  • Figure 30.2.png
    261.9KB
  • Figure 30.6.png
    335.7KB
  • Figure 30.15.png
    97.8KB
  • Figure 30.3.png
    328.4KB
  • segment.mat
    599B
  • normalise_structural.mat
    459B
  • readme.md
    1.3KB
  • Figure 30.19.png
    436KB
  • normalise_functional.mat
    1.5KB
  • realign.mat
    825B
  • Figure 30.13.png
    104.1KB
  • estimate.mat
    344B
  • coregister.mat
    478B
  • Figure 30.7.png
    446.8KB
  • specify.mat
    2.2KB
  • Figure 30.5.png
    300.1KB
  • smooth.mat
    677B
  • Figure 30.8.png
    206.1KB
  • Python
  • brain_state_classification.py
    4.2KB
  • readme.md
    1.2KB
  • Histogram, Entropy and Segmentation.ipynb
    431.1KB
  • Face Segmentation and Detection.ipynb
    161.3KB
  • MNIST-CNN
  • readme.md
    74B
  • mnist_cnn.py
    2.1KB
  • Iris_Recognition.py
    13.9KB
  • README.md
    3KB
内容介绍
# Image-Analysis: Exploring Image Analysis and its Applications ## Face Detection Detect a face in an image using k-means clustering based on color. ## Iris Recognition Implementation of John Daugman's Iris Detection algorithm and Li Ma's spatial filters to detect and recognise iris. ### 1. Iris Localization Obtain the exact parameters of the pupil and iris circles using Canny edge detection and Hough transform in a certain region determined by the center of the pupil. Return the coordinate and radius information for both iris and pupil. ### 2. Iris Normalization Map the iris from Cartesian coordinates to polar coordinates using the method provided in Li Ma's paper. Return a normalized image with dimensions 68 x 512. ### 3. Image Enhancement The normalized iris image has low contrast and may have nonuniform brightness caused by the position of light sources. We enhance the image by means of histogram equalization in each 32 x 32 region of the image. Li Ma's Region of Interest consists of only 48 x 512 part of the image so we return the enhanced image of only this ROI. ### 4. Feature Extraction Gabor filters have been used to filter the image and 2D convolution has been used to convolve the original image with the filter to obtain the features in the image. To characterize local texture information of the iris, statistical features in each 8 x 8 small block of the two filtered images have been extracted. In total we have 768 blocks and for each small block, two feature values are captured. This generates 1,536 feature components. The feature values used in the algorithm are the mean and the average absolute deviation of the magnitude of each filtered block. ### 5. Iris Matching Linear Discriminant Analysis is used to reduce the dimensionality of the high-dimensionality dataset. Eigen value decomposition is used as the solver, the shrinkage is automatic and the number of components has been seen in a range from 1 to the 108. After getting the reduced dataset, the model is trained using k nearest neighbors where the metrics being compared are 'cosine similarity' and 'manhattan distance'. The knn model learns from the training set and makes predictions based on the information it learns. The accuracy of the model is calculated in terms of the correct recognition rate (how many matches were correct), and this value is ~62.27% for cosine similarity and ~61.34% for Manhattan distance. A check was also performed to see that the values of the model's performances converge at these values. ### 6. Limitations There is still some imprecision as the eyelashes and the image noise have not been completely dealt with. ## Brain State Classification 1. Using the MATLAB sofware SPM to identify the brain regions that repond to auditory signals based on brain images. 2. Using Python for brain state classification challenges. ## MNIST Handwritten Digit Recognition using Convolutional Neural Networks Building a classifier that learns to identify handwritten digits using convolutional neural networks.
评论
    相关推荐
    • K均值聚类算法
      模式识别中常用的聚类算法,采用控制台程序,输出聚类结果
    • 聚类算法程序
      聚类算法中的感知器算法,用于模式识别中,有比较完整的算法描述
    • 谱聚类 聚类算法
      谱聚类 聚类算法 spectralClustering
    • k均值聚类算法
      模式识别作业之k均值聚类算法
    • 图像聚类算法图像聚类算法
      图像聚类算法图像聚类算法图像聚类算法图像聚类算法图像聚类算法图像聚类算法
    • Ncut聚类算法
      Ncut聚类算法,可以直接运行,有例子,程序有注释。
    • DBSCAN聚类算法
      经典DBASCAN聚类算法,适合新手小白学习,提供了数据可出效果图
    • ap聚类算法
      一个很好的利用DTW距离作相似度的ap聚类算法,同时能够自适应的调整最佳参数
    • DBSCAN聚类算法
      利用经典的基于密度的聚类算法,将四线激光雷达采集的数据进行聚类,剔除干扰点
    • AP聚类算法代码
      给出了AP聚类算法的实现代码,并给出了一个对二维坐标点进行聚类的实际例子的聚类结果。