import cv2
import numpy as np
import matplotlib.pyplot as plt
points = np. random. randint(0,100, (20,2))                        #隨機選擇20個點
labels = np. random. randint(0,2,(20,1))                           #為隨機點隨機分配標(biāo)志
label0s = points[labels.ravel()==0]                                #分出標(biāo)志為0的點
plt. scatter(label0s[:,0],label0s[:,1],80,'b','s')                 #將標(biāo)志為0的點繪制為藍(lán)色矩形
label1s = points[labels.ravel()==1]                                #分出標(biāo)志為1的點
plt. scatter(label1s[:,0],label1s[:,1],80,'r','^')                 #將標(biāo)志為1的點繪制為紅色三角形
newpoint = np.random.randint(0,100,(1,2))                          #隨機選擇一個點，下面確定其分類
plt.scatter(newpoint[:,0],newpoint[:,1],80,'g','o')                #將待分類新點繪制為綠色圓點
plt.show()                                                         #進一步使用kNN算法確認(rèn)待分類新點的類別、3個最近鄰居和距離
knn = cv2.ml.KNearest_create()                                     #創(chuàng)建kNN分類器
knn.train(points.astype(np.float32).cv2.ml.ROW_SAMPLE,labels.astype(np. float32))  #訓(xùn)練模型
ret,results,neighbours,dist = knn.findNearest(newpoint.astype(np.float32), 3)      #找出3個最近鄰居
print("新點標(biāo)志: %s" % results)
print("鄰居: %s" % neighbours)
print("距離: %s" % dist)

import cv2
import numpy as np
import matplotlib.pyplot as plt
gray = cv2. imread('img/a.png',0)     #讀入手寫數(shù)字的灰度圖像
digits = [np.hsplit(r,100) for r in np.vsplit(gray,50)]    #分解數(shù)字: 50行、100 列
np_digits = np.array(digits )      #轉(zhuǎn)換為NumPy數(shù)組
#準(zhǔn)備訓(xùn)練數(shù)據(jù)，轉(zhuǎn)換為二維數(shù)組，每個圖像400個像素
train_data = np_digits.reshape(-1, 400). astype(np. float32)
train_labels = np.repeat(np. arange(10),500)[:,np.newaxis]    #定義標(biāo)志
knn = cv2.ml.KNearest_create()             #創(chuàng)建kNN分類器
knn.train(train_data, cv2.ml.ROW_SAMPLE,train_labels)     #訓(xùn)練模型
#用繪圖工具創(chuàng)建的手寫數(shù)字5圖像(大小為20x 20 )進行測試
test= cv2.imread('img/a5.png',0)        #打開圖像
test_data=test.reshape(1,400).astype(np.float32)     #轉(zhuǎn)換為測試數(shù)據(jù)
ret,result,neighbours,dist = knn. findNearest(test_data,k=3)     #執(zhí)行測試
print(result.ravel())        #輸出測試結(jié)果
print(neighbours.ravel())
#將對手寫數(shù)字9拍攝所得圖像的大小轉(zhuǎn)換為20 x 20進行測試
img2=cv2. imread('img/d3.png',0)
ret,img2=cv2. threshold(img2,150, 255, cv2. THRESH_BINARY_INV)     #反二值化閾值處理
test_data=img2.reshape(1,400).astype(np. float32)       #轉(zhuǎn)換為測試數(shù)據(jù)
ret,result,neighbours,dist = knn. findNearest(test_data,k=3)                 #執(zhí)行測試
print(result.ravel())                                                               #輸出測試結(jié)果
print(neighbours .ravel())

import cv2
import numpy as np
import matplotlib.pyplot as plt#準(zhǔn)備訓(xùn)練數(shù)據(jù)，假設(shè)圖像高240,寬320,在其中選擇5個點
traindata=np.matrix([[140,60],[80, 120],[16,10],[166,190],[248, 180]],dtype = np.float32)
#5個點，前3個點為一類，標(biāo)志為8;后2個點為一類，標(biāo)志為1
labels = np.array([0,0,0,1,1])
svm = cv2.ml.SVM_create()             #創(chuàng)建SVM分類器
svm. setGamma(0.50625)                 #設(shè)置相關(guān)參數(shù)
svm. setC(12.5)
svm. setKernel(cv2.ml.SVM_LINEAR)
svm.setType(cv2.ml.SVM_C_SVC)
svm. setTermCriteria((cv2. TERM_CRITERIA_MAX_ITER, 100, 1e-6))
svm.train(traindata, cv2.ml. ROW_SAMPLE, labels)        #訓(xùn)練模型
img = np.zeros((240,320,3),dtype="uint8" )           #創(chuàng)建圖像
colors = {0:(102,255,204),1:(204,204,102)}
#用SVM分類器對圖像像素進行分類，根據(jù)分類結(jié)果設(shè)置像素顏色
for i in range(240):
    for j in range(320):
        point = np.matrix([[j,i]], dtype=np.float32)       #將像素坐標(biāo)轉(zhuǎn)換為測試數(shù)據(jù)
        label = svm.predict(point)[1].ravel()              #執(zhí)行預(yù)測，返回結(jié)果
        img[i,j] = colors[label[0]]              #根據(jù)預(yù)測結(jié)果設(shè)置像素顏色
svm_vectors = svm. getUncompressedSupportVectors() #獲得SVM向量
for i in range(svm_vectors.shape[0]):           #在圖像中繪制SVM向量(紅色圓)
    cv2.circle(img,(int(svm_vectors[i,0]),int(svm_vectors[i,1])),8,(0,0,255),2)
#在圖像中繪制訓(xùn)練數(shù)據(jù)點，類別標(biāo)志0使用藍(lán)色，類別標(biāo)志1使用綠色
cv2.circle(img, (140, 60),5,(255,0,0),-1)
cv2.circle(img, (80,120),5,(255,0,0),-1)
cv2.circle(img, (160,110),5,(255,0,0),-1)
cv2.circle(img, (160,190),5,(0,255,0),-1)
cv2.circle(img, (240,180),5,(0,255,0),-1)
img = cv2. cvtColor(img, cv2.COLOR_BGR2RGB)   #轉(zhuǎn)換為RGB格式
plt. imshow(img)
plt. show()        #顯示結(jié)果

import cv2
import numpy as np
def hog(img): #定義HOG描述符的計算函數(shù)
    hog = cv2.HOGDescriptor((20,20),(8,8),(4,4),(8,8),9,1,-1,0,0.2,1,64,True)    #定義HOGDescriptor對象
    hog_descriptor=hog.compute(img)            #計算HOG描述符
    hog_descriptor=np.squeeze(hog_descriptor)  #轉(zhuǎn)換為一維數(shù)組
    return hog_descriptor                       #返回HOG描述符，144 位
img = cv2.imread('img/a.png',0)
digits=[np.hsplit(row,100) for row in np.vsplit(img, 50)]  #分解圖像，50行、100列
labels = np.repeat(np. arange(10),500)[:,np.newaxis]   #定義對應(yīng)的標(biāo)記
hogdata = [list(map(hog,row)) for row in digits]        #計算圖像的HOG描述符
trainData = np. float32(hogdata) . reshape(-1, 144)    #轉(zhuǎn)換為測試數(shù)據(jù)
svm = cv2.ml.SVM_create( )      #創(chuàng)建SVM分類器
#設(shè)置相關(guān)參數(shù)
svm. setKernel(cv2.ml. SVM_LINEAR)
svm. setType(cv2.ml.SVM_C_SVC)
svm. setC(2.67)
svm. setGamma(5.383)
svm. train(trainData, cv2.ml. ROW_SAMPLE, labels)
#訓(xùn)練模型
#用繪圖工具創(chuàng)建的手寫數(shù)字5圖像(大小為20 x 20 )進行測試
test= cv2.imread('img/d5.jpg',0)
#打開圖像
test_data=hog(test)
test_data=test_data.reshape(1, 144) . astype(np. float32)
#轉(zhuǎn)換為測試數(shù)據(jù)
result = svm. predict(test_data)[1]
print( '識別結(jié)果: ',np. squeeze(result))
#用繪圖工具創(chuàng)建的手寫數(shù)字8圖像(大小為20 x20 )進行測試
test= cv2. imread('img/d7.jpg' ,0)
test_data=hog(test)
test_data=test_data.reshape(1,144).astype(np .float32)
#轉(zhuǎn)換為測試數(shù)據(jù)
result = svm. predict(test_data)[1]
print('識別結(jié)果:',np.squeeze(result))

import cv2
import numpy as np
from matplotlib import pyplot as plt
#創(chuàng)建聚類數(shù)據(jù)，3個類別，每個類別包含20個點
data = np.vstack((np.random. randint(10,90, (20,2)),np. random. randint(80,170, (20, 2)),np.random.randint(160,250, (20, 2))))
data=data.astype(np.float32)
#定義算法終止條件
criteria = (cv2.TERM_CRITERIA_EPS + cv2.TERM_CRITERIA_MAX_ITER, 20, 1.0)
#使用k均值聚類算法執(zhí)行分類操作，k=3，返回結(jié)果中l(wèi)abel用于保存標(biāo)志，center 用于保存質(zhì)心
ret,label,center=cv2. kmeans(data,3,None, criteria,10,cv2.KMEANS_RANDOM_CENTERS)
#根據(jù)運算結(jié)果返回的標(biāo)志將數(shù)據(jù)分為3組，便于繪制圖像
data1 = data[label.ravel() == 0]
data2 = data[label.ravel() == 1]
data3 = data[label.ravel() == 2]
plt. scatter(data1[:,0], data1[:,1], c='r')
#繪制第1類數(shù)據(jù)點，紅色
#繪制第2類數(shù)據(jù)點，綠色
plt.scatter(data2[:,0],data2[:,1],c='g')
#繪制第3類數(shù)據(jù)點，藍(lán)色
plt.scatter(data3[:,0], data3[:,1], c='b')
plt.scatter(center[:,0],center[:,1],100,['#CC3399'],'s')
#繪制質(zhì)心，顏色為#CC3399
#顯示結(jié)果
plt. show()

import cv2
import numpy as np
img = cv2.imread('img/qwe.jpg')
#打開圖像

#顯示原圖
img2 = img.reshape((-1,3)). astype(np. float32)
#轉(zhuǎn)換為nx 3的浮點類型數(shù)組，n=圖像像素的總數(shù)一3
#定義算法終止條件
criteria = (cv2.TERM_CRITERIA_EPS + cv2.TERM_CRITERIA_MAX_ITER, 10, 1.0)
K=8
ret,label,center=cv2.kmeans(img2,K,None, criteria,10,cv2.KMEANS_RANDOM_CENTERS)
center = np.uint8(center)             #將質(zhì)心轉(zhuǎn)換為整型
img3 = center[label.ravel()]          #轉(zhuǎn)換為一維數(shù)組
img3 = img3.reshape((img.shape))        #恢復(fù)為原圖像數(shù)組形狀
cv2.imshow('K=8',img3)
K=12
ret,label,center=cv2. kmeans(img2,K,None, criteria,10,cv2.KMEANS_RANDOM_CENTERS)
center = np.uint8(center)             #將質(zhì)心轉(zhuǎn)換為整型
img3 = center[label.ravel()]          #轉(zhuǎn)換為一維數(shù)組
img3 = img3.reshape((img.shape))        #恢復(fù)為原圖像數(shù)組形狀
cv2.imshow('K=12',img3)
cv2.waitKey(0)

import cv2
import numpy as np
from matplotlib import pyplot as plt
from PIL import ImageFont,ImageDraw, Image
#讀入文本文件中的類別名稱，共0種類到每行為一個類到，從第11個字符開始為名稱
#基本格式如下
# n01440764 tench, Tinca tinca
# n01443537 goldfish, Carassius auratus
file=open('classes.txt' )
names=[r.strip() for r in file.readlines()]
file.close()
classes = [r[10:] for r in names]
#獲取每個類別的名稱
#從文件中載入Caffe模型
net = cv2.dnn.readNetFromCaffe("AlexNet_deploy.txt", "AlexNet_CaffeModel.dat")
image = cv2. imread("img/qwe.jpg")
#打開圖像，用于識別分類
#創(chuàng)建圖像塊數(shù)據(jù)，大小為(224,224)，顏色通道的均值縮減比例因子為(104, 117, 123)
blob = cv2.dnn. blobFromImage(image, 1, (224,224), (104, 117, 123))
net. setInput(blob)
#將圖像塊數(shù)據(jù)作為神經(jīng)網(wǎng)絡(luò)輸入
#執(zhí)行預(yù)測，返回結(jié)果是一個 1x 1000的數(shù)組，按順序?qū)?yīng)1000種類別的可信度
result = net. forward()
ptime, x = net. getPerfProfile()
#獲得完成預(yù)測時間
print('完成預(yù)測時間: %.2f ms' % (ptime * 1000.0 / cv2.getTickFrequency(0)))
sorted_ret = np. argsort(result[0])
#將預(yù)測結(jié)果按可信度由高到低排序
top5 = sorted. ret[::-1][:5]
#獲得排名前5的預(yù)測結(jié)果
ctext = "類別: "+classes[top5[0]]
ptext = "可信度: {:.2%}" . format(result[0][top5[0]])
#輸出排名前5的預(yù)測結(jié)果
for (index, idx) in enumerate(top5):
    print("t).類別: {},可信度: (:2%)" .format(index + 1, classes[idx], result[0][idx]))
#在圖像中輸出排名第1的預(yù)測結(jié)果
fontpath = "STSONG. TTF"
font = ImageFont . truetype( fontpath, 80)        #載入中文字體，設(shè)置字號
img_pil = Image. fromarray(image)
draw = ImageDraw.Draw(img_pil)
draw.text((10, 10),ctext, font = font,fill=(0,0,255)) #繪制文字
draw.text((10,100),ptext, font = font, fill=(0,0,255))
img = np.array(img_pil)
img = cv2.cvtColor(img,cv2.COLOR_BGR2RGB)
plt.imshow(img)
plt.axis('off')
plt.show()

import cv2
import numpy as np
from matplotlib import pyplot as plt
import matplotlib
from PIL import InageFont, Inagoraw, Image
#加載字體，以便顯示漢字
fontpath = "STSONG.TTF”
font = ImageFont.truetype(fontpath,20)     #載入字體，設(shè)置字號
font2 = {'family': 'STSONG', "size": 22}
matplotlib.rc('font', **font2)
#設(shè)置plt字體
#從文件中加載已知的對象名稱，文件保存了80個類別的對象名稱，每行個
f=open(" object_names .txt",encoding=' utf-8')
object. names = [r.strip() for r in f.readlines()]
f.close()
#從文件中加載預(yù)訓(xùn)練的Darknet模型
mode = cv2. dnn. readNetFromDarknet("yolov3.cfg", "yolov3.weights")
image = cv2. imread(" objects. jpg")
#打開圖像文件
imgH, imgW = image . shape[ :2]
ut_layers = mode. getLayerNames()
#獲得輸出層
out_layers = [out_layers[i[0] - 1] for i in mode.getUnconnectedOutLayers()]
blob = cv2.dnn. blobFromImage(image, 1/255.0, (416,416),swapRB=True, crop=False)  #創(chuàng)建圖像塊數(shù)據(jù)
mode . setInput(blob)
#將圖像塊數(shù)據(jù)設(shè)置為模型輸入
layer_results = mode . forward(out_layers)
#執(zhí)行預(yù)測，返回每層的預(yù)測結(jié)果
ptime,_ = mode. getPerfProfile()
tilte_text='完成預(yù)測時間: %.2f ms' % (ptime* 1000/cv2. getTickFrequency())
result_boxes = []
result_scores = []
result_name_id = []
for layer in layer_results:        #遍歷所有輸出層
    for box in layer:              #遍歷層的所有輸出預(yù)測結(jié)果，每個結(jié)果為一個邊框
        #預(yù)測結(jié)果結(jié)構(gòu): x, y, w, h, confidence, 80 個類別的概率
        probs = box[5:]
        class_id = np.argmax(probs)     #找到概率最大的類別名稱
        prob = probs[class_id]          #找到最大的概率
        if prob > 0.5:
            #篩選出概率大于50%的類別
            #計算每個box在原圖像中的絕對坐標(biāo)
            box = box[0:4]* np.array[imgW, imgH, imgW, imgH])
            (centerx, centery, width, heignt) = box.astype("int")
            x = int(centerX - (width / 2))
            y = int(centerY - (height / 2))
            result_boxes.append([x, y, int(width),int(height)])
            result_scores . append(float(prob))
            result_name_id. append(class_id)
#應(yīng)用非最大值抑制消除重復(fù)邊框，獲得要繪制的box
draw_boxes = cv2.dnn. NMSBoxes(result_boxes, result_scores, 0.6, 0.3)
if len(draw_boxes) > 0:
    for i in draw_boxes.ravel():
        #獲得邊框坐標(biāo)
        (x, y) = (result_boxes[i][0], result_boxes[i][1])
        (w, h) = (result_boxes[i][2], result_boxes[i][3])
        #繪制邊框
        cv2.rectangle(image,(x,y), (x+w,y+h),(0,255,0),1)
        #輸出類別名稱和可信度
        text=object_names[result_name_id[i]] +"\n{: .1%}". format(result_scores[i])
        img_pil = Image.fromarray(image)
        draw = ImageDraw.Draw(img_pil)
        draw.text((x+5,y), text, font = font,fill=(0,0,255))
        #繪制文字
        image = np.array(img_pil)
img = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
plt. title(tilte_text)
plt. imshow(img)
plt.axis('off')
plt. show()