机器学习大作业-python实现的掌纹识别项目源码（高分项目）.zip

import os import cv2 import numpy as np from sklearn.decomposition import PCA from sklearn import svm from sklearn.metrics import accuracy_score from skimage import feature from tqdm import tqdm from sklearn.preprocessing import MinMaxScaler,StandardScaler from sklearn.cluster import KMeans from sklearn.model_selection import GridSearchCV from joblib import dump from sklearn.decomposition import IncrementalPCA def get_img(path): img_list=[] label_list=[] for filename in os.listdir(path): label=int(filename.split('_')[1]) id=filename.split('_')[2].split('.')[0] img=cv2.imread(os.path.join(path, filename)) img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY) img_list.append(img) label_list.append(label) return np.array(img_list),np.array(label_list) def apply_gabor_filter(image, kernel): filtered_image = cv2.filter2D(image, cv2.CV_8UC3, kernel) return filtered_image def Gabor(image): gabor_filters = [] # 读取图像 # image = cv2.cvtColor(rgb_image, cv2.COLOR_BGR2GRAY) # 定义Gabor滤波器参数 gabor_size = [6,9,12,15,18,21] # 滤波器大小 sigma = 4.0 # 高斯核标准差 lamda = np.pi/1.0 #波长 gamma = 0.5 # 空间纵横比 # 创建Gabor滤波器 # kernel = cv2.getGaborKernel(ksize, sigma, theta, lambd, gamma) for theta in np.arange(0,np.pi,np.pi / 4): #定义gabor的4个方向 for i in range(6): kern = cv2.getGaborKernel((gabor_size[i],gabor_size[i]),1.0,theta,lamda,0.5,0,ktype=cv2.CV_32F) kern /= 1.2*kern.sum() gabor_filters.append(kern) # 应用Gabor滤波器 res = [] #滤波结果 for i in range(len(gabor_filters)): res1 = apply_gabor_filter(image,gabor_filters[i]) res.append(np.asarray(res1)) return np.array(res) def feature_view(sobelx,sobely,lbp,res): # 先将各特征进行展平,再进行归一化，然后拼接。 # 创建MinMaxScaler对象 scaler = MinMaxScaler() r = res.reshape(-1, 1) r=scaler.fit_transform(r) r=np.transpose(r,(1,0)) s = np.stack((sobelx, sobely)).reshape(-1, 1) s=scaler.fit_transform(s) s=np.transpose(s,(1,0)) l=lbp.reshape(-1, 1) l=scaler.fit_transform(l) l=np.transpose(l,(1,0)) # s=scaler.fit_transform(s) # r=scaler.fit_transform(r) # l=scaler.fit_transform(l) #水平拼接 result_feature = np.hstack((r,s,l)) return result_feature def preprocessing(image): # 直方图均衡化 image = cv2.equalizeHist(image) # 应用Sobel边缘检测 sobelx = cv2.Sobel(image, cv2.CV_64F, 1, 0, ksize=3)#卷积核大小 sobely = cv2.Sobel(image, cv2.CV_64F, 0, 1, ksize=3) # 应用LBP特征提取 radius = 1 # LBP算法中范围半径的取值 n_points = 8 * radius # 领域像素点数 lbp = feature.local_binary_pattern(image, n_points, radius, method="uniform") # Gabor滤波器 res=Gabor(image) result_feature=feature_view(sobelx,sobely,lbp,res) return result_feature def get_sift(image): # 直方图均衡化 image = cv2.equalizeHist(image) # 创建SIFT对象 sift = cv2.xfeatures2d.SIFT_create() # 检测关键点和计算描述符 keypoints, descriptors = sift.detectAndCompute(image, None) # 在图像上绘制关键点 return descriptors def BoW_sift(X_train,k=100): descriptors_list=[] for img in X_train: descriptors=get_sift(img) descriptors_list.append(descriptors) # 将所有图像的SIFT特征放在一起 all_sift_features=np.vstack(descriptors_list) # 使用K-means聚类创建视觉词典 # k = 100 # 假设我们想要的视觉单词数量是100 kmeans = KMeans(n_clusters=k) kmeans.fit(all_sift_features) # 对每张图像，计算每个视觉单词的出现次数 image_features = [] for sift_features in descriptors_list: labels = kmeans.predict(sift_features) hist, _ = np.histogram(labels, bins=np.arange(k+1)) image_features.append(hist) return image_features,kmeans def test_BoW_sift(X_test,kmeans,k=100): descriptors_list = [] for img in X_test: descriptors = get_sift(img) descriptors_list.append(descriptors) image_features = [] for sift_features in descriptors_list: labels = kmeans.predict(sift_features) hist, _ = np.histogram(labels, bins=np.arange(k + 1)) image_features.append(hist) return image_features def first_feature(X_train): # 初始化一个空的列表来保存所有的特征向量 features = [] # 循环处理每一张图片 for image in tqdm(X_train): # 对图片进行预处理并获取特征向量 result_feature = preprocessing(image) # 将特征向量添加到列表中 features.append(result_feature) # 使用np.concatenate()将所有的特征向量拼接在一起 first_features = np.concatenate(features, axis=0) return first_features def temp_feature(X_train): # 初始化一个空的数组来保存所有的特征向量 first_features = np.array([]) # 循环处理每一张图片 for image in tqdm(X_train): # 对图片进行预处理并获取特征向量 result_feature = preprocessing(image) print(result_feature.shape) # 将特征向量添加到数组中 first_features = np.concatenate((first_features, result_feature), axis=0) # 保存特征向量到硬盘 np.save('temp_data/first_features.npy', first_features) def PCA_trainfeature(X_train,n_components=1000): first_features=first_feature(X_train) # 拟合PCA模型 # n_components = 1000 pca = PCA(n_components=n_components) # 对数据进行降维 X_train_pca = pca.fit_transform(first_features) sift_features,kmeans=BoW_sift(X_train) # 使用dstack进行第三维度上的拼接 sift_array = np.dstack(sift_features)[0] scaler = MinMaxScaler() scaler_sift_array=scaler.fit_transform(sift_array) scaler_sift_array=np.transpose(scaler_sift_array,(1,0)) trainfeature=np.concatenate((X_train_pca,scaler_sift_array), axis=1) return trainfeature,pca,kmeans def get_testfeature(X_test,pca,kmeans): first_features=first_feature(X_test) # 使用相同的PCA对象对测试数据进行降维 X_test_pca = pca.transform(first_features) sift_features=test_BoW_sift(X_test,kmeans) # 使用dstack进行第三维度上的拼接 sift_array = np.dstack(sift_features)[0] scaler = MinMaxScaler() scaler_sift_array=scaler.fit_transform(sift_array) scaler_sift_array=np.transpose(scaler_sift_array,(1,0)) testfeature=np.concatenate((X_test_pca,scaler_sift_array), axis=1) return testfeature def batch_get_feature(first_features,pca,kmeans,X_test): X_test_pca = pca.transform(first_features) sift_features=test_BoW_sift(X_test,kmeans) # 使用dstack进行第三维度上的拼接 sift_array = np.dstack(sift_features)[0] scaler = MinMaxScaler() scaler_sift_array=scaler.fit_transform(sift_array) scaler_sift_array=np.transpose(scaler_sift_array,(1,0)) testfeature=np.concatenate((X_test_pca,scaler_sift_array), axis=1) return testfeature def batch_feature(X_train,pca,kmeans,batch): # 初始化一个空的数组来保存所有的特征向量 # result_features = np.empty((0, 442368)) pre_features = np.empty((0, 500)) # 初始化一个空的列表来保存每一批次的特征向量 batch_features = [] image_list=[] # 循环处理每一张图片 for i,image in enumerate(tqdm(X_train)): # 对图片进行预处理并获取特征向量 image_feature = preprocessing(image) image_list.append(image) # 将特征向量添加到列表中 batch_features.append(image_feature) # 每迭代batch次，将特征向量添加到数组中，并清空列表