60. hog+svm小狮子识别

# 样本 训练 预测test
# 样本 pos:neg = 1:2 或1:3
# 步骤: 1参数  2hog  3svm 4compute hog  5label 6train 7pred 8draw

import cv2
import numpy as np
import matplotlib.pyplot as plt

# 1par
posNum = 820 
negNum = 1931
winSize = (64, 128)
blockSize = (16, 16) # 105block
blockStride = (8, 8) 
cellSize = (8, 8) # 每个block有4cell
nBin = 9 # 每个cell有9bin      => 3780维

# 2hog
hog = cv2.HOGDescriptor(winSize, blockSize, blockStride, cellSize, nBin)

# 3svm
svm =  cv2.ml.SVM_create()

# 4 comoute hog
featureNum = int(((128 - 16)/8 + 1) * ((64 - 16)/ 8 + 1)* 4 * 9) # 3780
print('featureNum:', featureNum)
featureArray = np.zeros((posNum + negNum, featureNum), np.float32)

# 5 label 
labelArray = np.zeros((posNum + negNum, 1), np.int32)

# svm监督学习需要样本（中的hog特征）和标签
for i in range(0, posNum):
    fileName = 'pos/' + str(i + 1) + '.jpg'
    img = cv2.imread(fileName)
    hist = hog.compute(img, (8, 8)) # 3780
    
    # hist(3780维)装入featureArray
    for j in range(0, featureNum):
        featureArray[i, j] = hist[j]
        # hog1: hog[1, :]   hog2: hog[2, :]
        # n*1 第一个是正样本  正样本处理完
    labelArray[i, 0] = 1
    
for i in range(0, negNum):
    fileName = 'neg/' + str(i + 1) + '.jpg'
    img = cv2.imread(fileName)
    hist = hog.compute(img, (8, 8))
    
    for j in range(0, featureNum):
        featureArray[i + posNum, j] = hist[j]
    # 负样本处理完
    labelArray[i + posNum, 0] = -1
    
svm.setType(cv2.ml.SVM_C_SVC)
svm.setKernel(cv2.ml.SVM_LINEAR)
svm.setC(0.01)


# 6train
ret = svm.train(featureArray, cv2.ml.ROW_SAMPLE, labelArray)

# 一维
alpha = np.zeros((1), np.float32)
# svm训练后得到，判决时用到
rho = svm.getDecisionFunction(0, alpha)
print('alpha: ', alpha)
print('rho: ', rho)

# 与svm数组相乘  alphaArray为一行一列的二维数组【一维】
alphaArray = np.zeros((1, 1), np.float32)
supportVArray = np.zeros((1, featureNum), np.float32)  # 支持向量个数
resultArray = np.zeros((1, featureNum), np.float32)

alphaArray[0, 0] = alpha 
resultArray = -1 * alphaArray * supportVArray  # 3780维


# 7 test（detect）   核心：myHog-> myDetect
# myDetect为一维数组，来自resultArray与rho；最后myHog.detectMultiScale
myDetect = np.zeros((3781), np.float32)  
for i in range(0, 3780): 
    myDetect[i] = resultArray[0, i]
    
myDetect[3780] = rho[0]


# 构建hog
myHog = cv2.HOGDescriptor()
myHog.setSVMDetector(myDetect)

# load 
imageSrc = cv2.imread('Test.jpg', 1)
# detectMultiScale是的检测核心  win滑动步长:(8, 8)  winSize:(32, 32)  缩放系数：1.05

objs = myHog.detectMultiScale(imageSrc, 0, (8, 8), (32, 32), 1.05, 2)
# xy wh 三维 最后一维的0-3
x = int(objs[0][0][0])
y = int(objs[0][0][1])
w = int(objs[0][0][2])
h = int(objs[0][0][3])

# 8 绘制展示
cv2.rectangle(imageSrc, (x, y), (x + w, y + h), (255, 0, 0), 2)
cv2.imshow('dst', imageSrc)

cv2.waitKey(0) 

识别结果如下：

image.png