From ec97d7470529b2a5c62447feebb15ffedfa2a4d2 Mon Sep 17 00:00:00 2001 From: coral Date: Fri, 1 Jul 2022 11:56:25 +0800 Subject: [PATCH] image processiong; LR; cls --- .DS_Store | Bin 0 -> 6148 bytes Classifer.py | 67 +++++++++++++++++++++++++++++ LR.py | 52 +++++++++++++++++++++++ demo_cls.py | 8 ++++ demo_img.py | 9 ++++ demo_reg.py | 8 ++++ mmImage.py | 118 +++++++++++++++++++++++++++++++++++++++++++++++++++ 7 files changed, 262 insertions(+) create mode 100644 .DS_Store create mode 100644 Classifer.py create mode 100644 LR.py create mode 100644 demo_cls.py create mode 100644 demo_img.py create mode 100644 demo_reg.py create mode 100644 mmImage.py diff --git a/.DS_Store b/.DS_Store new file mode 100644 index 0000000000000000000000000000000000000000..6ec67e066cb641a83a7481fcadabc62fa0018939 GIT binary patch literal 6148 zcmeHK%}T>S5dKz+)~eW>UW7b>;K^&M$wB;~TJLI`QiawO(SsnEyZ8#ef=}Sh2k}kx zo83i{H1#N=GiCSN?9R8-`5-%40A~9(*#w#Z>TH6w2D=KAb!ihe<7bwLM)nvEHaop^ zG@kI(2~|K9_>Bt4yIaEs2Drpc@%?>z&99BpDC!Q=F6Zj2&HIO=t+!TjJ^6Nhe33gF z%sZ)=#|~OJ#tGw(ako26dp?33N3#3Ihc$7=9QJU?UzP7m$#)+6*yZjq{X2+@JDc^A zcLO5~af$?ej?$_K=CQ0m^tJM%{VI2QH{T17)R%P;Qd06nL|g1@s|(d zBO8B1F*Z8)53D;(=umrAKouw|FzX(hQvc)W_x~bEpHu-=;7=)FYEdWJW=g)cP9!I_ tHefqr6BED8p{20%<=76S6tA&a<60mCV(2k*$Qhb`2v`}kQw4rhfln>6krV&` literal 0 HcmV?d00001 diff --git a/Classifer.py b/Classifer.py new file mode 100644 index 0000000..5c66594 --- /dev/null +++ b/Classifer.py @@ -0,0 +1,67 @@ +import pandas as pd +import numpy as np +import os + +from sklearn.metrics import accuracy_score ,mean_squared_error, r2_score +from sklearn.svm import SVC +from sklearn import linear_model +from sklearn.naive_bayes import GaussianNB +from sklearn.ensemble import BaggingClassifier,AdaBoostClassifier +from sklearn.neighbors import KNeighborsClassifier + +class Classifer: + def __init__ (self, + backbone='RandomForest' + ): + self.backbone = backbone #获取外部运行py的绝对路径 + self.cwd = os.path.dirname(os.getcwd()) #获取当前文件的绝对路径 + self.file_dirname = os.path.dirname(os.path.abspath(__file__)) + if backbone == 'SVM': + self.model = SVC() + elif backbone == 'NB': + self.model = GaussianNB() + elif backbone == 'bagging': + self.model = BaggingClassifier(KNeighborsClassifier()) + elif backbone == 'boosting': + self.model = AdaBoostClassifier(n_estimators=100, random_state=0) + + def train(self,seed=0,data_type='csv'): + np.random.seed(seed) + if data_type == 'csv': + dataset = pd.read_csv(self.dataset_path,sep=',',header=None).values + elif data_type == 'pandas': + dataset = self.load_pd() + elif data_type == 'list': + dataset = self.load_list() + np.random.shuffle(dataset) + + data,label = dataset[:,:-1],dataset[:,-1] + train_index = int((1-self.test_size)*len(dataset)) + train_data,train_label = data[:train_index,],label[:train_index] + self.test_set = { + 'data':data[train_index:,], + 'label':label[train_index:] + } + self.model.fit(train_data,train_label) + + def inference(self,mode = 'cls'): + pred = self.model.predict(self.test_set['data']) + if mode == 'cls': + acc = accuracy_score(self.test_set['label'],pred) + print('准确率为：{}%'.format(acc*100)) + elif mode == 'reg': + loss = mean_squared_error(self.test_set['label'],pred) + print('Loss: {}'.format(loss)) + + + + + def load_dataset(self,path,test_size=0.2): + self.dataset_path = path + self.test_size=test_size + + def load_pd(): + pass + + def load_list(): + pass \ No newline at end of file diff --git a/LR.py b/LR.py new file mode 100644 index 0000000..28f1291 --- /dev/null +++ b/LR.py @@ -0,0 +1,52 @@ +from turtle import back +import pandas as pd +import numpy as np +import os +import cv2 +import os +from sklearn.metrics import accuracy_score ,mean_squared_error, r2_score +from sklearn import linear_model + +class LR: + def __init__ (self, + backbone='LR' + ): + self.backbone = backbone #获取外部运行py的绝对路径 + self.cwd = os.path.dirname(os.getcwd()) #获取当前文件的绝对路径 + self.file_dirname = os.path.dirname(os.path.abspath(__file__)) + self.model = linear_model.LinearRegression() + + def train(self,seed=0,data_type='csv'): + np.random.seed(seed) + if data_type == 'csv': + dataset = pd.read_csv(self.dataset_path,sep=',',header=None).values + elif data_type == 'pandas': + dataset = self.load_pd() + elif data_type == 'list': + dataset = self.load_list() + np.random.shuffle(dataset) + + data,label = dataset[:,:-1],dataset[:,-1] + train_index = int((1-self.test_size)*len(dataset)) + train_data,train_label = data[:train_index,],label[:train_index] + self.test_set = { + 'data':data[train_index:,], + 'label':label[train_index:] + } + self.model.fit(train_data,train_label) + + def inference(self,mode = 'cls'): + pred = self.model.predict(self.test_set['data']) + loss = mean_squared_error(self.test_set['label'],pred) + print('Loss: {}'.format(loss)) + + + def load_dataset(self,path,test_size=0.2): + self.dataset_path = path + self.test_size=test_size + + def load_pd(): + pass + + def load_list(): + pass \ No newline at end of file diff --git a/demo_cls.py b/demo_cls.py new file mode 100644 index 0000000..31b319b --- /dev/null +++ b/demo_cls.py @@ -0,0 +1,8 @@ +# from base import * +from Classifer import * +# from MMEdu import MMMlearing +dataset_path = "./test.csv" +model = Classifer(backbone ='SVM') +model.load_dataset(dataset_path) +model.train() +acc = model.inference() diff --git a/demo_img.py b/demo_img.py new file mode 100644 index 0000000..d0faa58 --- /dev/null +++ b/demo_img.py @@ -0,0 +1,9 @@ +from mmImage import * +image_path = "6.png" +# model = MMImage(method = 'contour') +# model = MMImage(method = 'canny') +# model = MMImage(method = 'blur') #para = ['mean',3] +model = MMImage(method = 'corner') # para = 0.01) +model.load_image(image_path) +image_out = model.process(save_path = './save.png',para = 0.1) +# diff --git a/demo_reg.py b/demo_reg.py new file mode 100644 index 0000000..9cea765 --- /dev/null +++ b/demo_reg.py @@ -0,0 +1,8 @@ +# from base import * +from LR import * +# from MMEdu import MMMlearing +dataset_path = "./test.csv" +model = LR() +model.load_dataset(dataset_path) +model.train() +acc = model.inference() diff --git a/mmImage.py b/mmImage.py new file mode 100644 index 0000000..ae7a945 --- /dev/null +++ b/mmImage.py @@ -0,0 +1,118 @@ +from copy import deepcopy +import cv2 +import os +import numpy as np + +class MMImage: + def __init__ (self,method='blur'): + self.method = method + self.img = None + + def process(self,save_path = '',para = []): + if save_path != '': + save_path = save_path + img_out = getattr(self, "_"+self.method)(para) + cv2.imwrite(save_path,img_out) + + def load_image(self,image_path): + img = cv2.imread(image_path,cv2.IMREAD_UNCHANGED) + self.img = img + + def _blur(self,para = []): + type_method,kernal = para + img_out = cv2.blur(self.img, (kernal, kernal)) #sum(square)/25 + return img_out + + def _contour(self,para=[]): + gray_img=cv2.cvtColor(self.img,cv2.COLOR_BGR2GRAY) + dep,img_bin=cv2.threshold(gray_img,128,255,cv2.THRESH_BINARY) + image_,contours=cv2.findContours(img_bin, mode=cv2.RETR_TREE, method=cv2.CHAIN_APPROX_SIMPLE) + to_write = self.img.copy() + # cv2.drawContours(img,contours,0,(0,0,255),3) + ret = cv2.drawContours(to_write,image_,-1,(0,0,255),3) + return ret + + def _hist(self,para=[]): + gray_img=cv2.cvtColor(self.img,cv2.COLOR_BGR2GRAY) + img = gray_img.astype(np.uint8) + return cv2.equalizeHist(img) + + def _watershed_contour(self,para=[]): + gray = cv2.cvtColor(self.img, cv2.COLOR_BGR2GRAY) # 转为灰度图像 + + # 查找和绘制图像轮廓 + Gauss = cv2.GaussianBlur(gray, (5,5), sigmaX=4.0) + grad = cv2.Canny(Gauss,50,150) + + grad, contours = cv2.findContours(grad, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) # 查找图像轮廓 + markers = np.zeros(self.img.shape[:2], np.int32) # 生成标识图像，所有轮廓区域标识为索引号 (index) + for index in range(len(contours)): # 用轮廓的索引号 index 标识轮廓区域 + markers = cv2.drawContours(markers, grad, index, (index, index, index), 1, 8, contours) + ContoursMarkers = np.zeros(self.img.shape[:2], np.uint8) + ContoursMarkers[markers>0] = 255 + + # 分水岭算法 + markers = cv2.watershed(self.img, markers) # 所有轮廓的像素点被标注为 -1 + WatershedMarkers = cv2.convertScaleAbs(markers) + # 用随机颜色填充分割图像 + bgrMarkers = np.zeros_like(self.img) + for i in range(len(contours)): + colorKind = [np.random.randint(0, 255), np.random.randint(0, 255), np.random.randint(0, 255)] + bgrMarkers[markers==i] = colorKind + bgrFilled = cv2.addWeighted(self.img, 0.67, bgrMarkers, 0.33, 0) + return cv2.cvtColor(bgrFilled, cv2.COLOR_BGR2RGB) + + def _watershed(self,para): + gray = cv2.cvtColor(self.img, cv2.COLOR_BGR2GRAY) # 转为灰度图像 + + # 图像的形态学梯度 + kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (5, 5)) # 生成 5*5 结构元 + grad = cv2.morphologyEx(gray, cv2.MORPH_GRADIENT, kernel) # 形态学梯度 + + # 阈值分割，将灰度图像分为黑白二值图像 + _, thresh = cv2.threshold(np.uint8(grad), 0.2*grad.max(), 255, cv2.THRESH_BINARY) + # 形态学操作，生成 "确定背景" 区域 + kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (3, 3)) # 生成 3*3 结构元 + opening = cv2.morphologyEx(thresh, cv2.MORPH_OPEN, kernel, iterations=2) # 开运算，消除噪点 + sure_bg = cv2.dilate(opening, kernel, iterations=3) # 膨胀操作，生成 "确定背景" 区域 + # 距离变换，生成 "确定前景" 区域 + distance = cv2.distanceTransform(opening, cv2.DIST_L2, 5) # DIST_L2: 3/5 + _, sure_fg = cv2.threshold(distance, 0.1 * distance.max(), 255, 0) # 阈值选择 0.1*max 效果较好 + sure_fg = np.uint8(sure_fg) + # 连通域处理 + ret, component = cv2.connectedComponents(sure_fg, connectivity=8) # 对连通区域进行标号，序号为 0-N-1 + markers = component + 1 # OpenCV 分水岭算法设置标注从 1 开始，而连通域编从 0 开始 + kinds = markers.max() # 标注连通域的数量 + maxKind = np.argmax(np.bincount(markers.flatten())) # 出现最多的序号，所占面积最大，选为底色 + markersBGR = np.ones_like(self.img) * 255 + for i in range(kinds): + if (i!=maxKind): + colorKind = [np.random.randint(0, 255), np.random.randint(0, 255), np.random.randint(0, 255)] + markersBGR[markers==i] = colorKind + # 去除连通域中的背景区域部分 + unknown = cv2.subtract(sure_bg, sure_fg) # 待定区域，前景与背景的重合区域 + markers[unknown == 255] = 0 # 去掉属于背景的区域 (置零) + # 分水岭算法标注目标的轮廓 + markers = cv2.watershed(self.img, markers) # 分水岭算法，将所有轮廓的像素点标注为 -1 + kinds = markers.max() # 标注连通域的数量 + + # 把轮廓添加到原始图像上 + imgWatershed = self.img.copy() + imgWatershed[markers == -1] = [0, 0, 255] # 将分水岭算法标注的轮廓点设为红色 + # print(self.img.shape, markers.shape, markers.max(), markers.min(), ret) + return cv2.cvtColor(markersBGR, cv2.COLOR_BGR2RGB) + + + def _canny(self,para=[100,200]): + return cv2.Canny(self.img,para[0],para[1]) + + def _corner(self,para = 0.01): + gray_img = cv2.cvtColor(self.img, cv2.COLOR_BGR2GRAY) + gray_img = np.float32(gray_img) + Harris_detector = cv2.cornerHarris(gray_img, 2, 3, 0.04) + dst = cv2.dilate(Harris_detector, None) + # 设置阈值 + thres = para*dst.max() + img_out = deepcopy(self.img) + img_out[dst > thres] = [255,0,0] + return img_out