Article Directory
1.SVM principle
2. Use the SVM handwritten data OCR
- In kNN we directly use the pixel gray scale value as a feature vector. This time we use a gradient direction histogram Histogram of Oriented Gradients (HOG) as a feature vector.
- Before calculating HOG we use the second moment of its image deskew (deskew) process.
Code shorthand:
- cv2.ml.SVM_create()
- svm.train()
- svm.save()
- svm.predict()
Actual:
second moment (1) using an image subjected to deskew (Deskew) Processing
def deskew(img):
m = cv2.moments(img) # 求矩
if abs(m['mu02']) < 1e-2:
return img.copy()
skew = m['mu11'] / m['mu02']
M = np.float32([[1, skew, -0.5 * SZ * skew], [0, 1, 0]])
img = cv2.warpAffine(img, M, (SZ, SZ), flags=affine_flags)
return img
(2) using a gradient direction histogram Histogram of Oriented Gradients HOG as a feature vector.
- Sobel derivative calculation image number X direction and the Y direction. Then the calculated gradient magnitude and direction for each pixel. This gradient is converted into 16-bit integers.
- The image is divided into four small blocks, calculate their orientation histogram (16 bin) for each of the small block, the gradient magnitude do weights.
- So that each small box will get a vector containing 16 members. 4 vectors 4 on the formation of small blocks of the image feature vector (containing 64 members)
def hog(img):
gx = cv2.Sobel(img, cv2.CV_32F, 1, 0)
gy = cv2.Sobel(img, cv2.CV_32F, 0, 1)
mag, ang = cv2.cartToPolar(gx, gy)
bins = np.int32(bin_n * ang / (2 * np.pi))# quantizing binvalues in (0...16)
bin_cells = bins[:10, :10], bins[10:, :10], bins[:10, 10:], bins[10:, 10:]
mag_cells = mag[:10, :10], mag[10:, :10], mag[:10, 10:], mag[10:, 10:]
hists = [np.bincount(b.ravel(), m.ravel(), bin_n) for b, m in zip(bin_cells, mag_cells)]
hist = np.hstack(hists)# hist is a 64 bit vector
return hist
(3) use of a digital handwriting recognition SVM
def opencv_svm(self):
img = self.img
# 【1】切割图像
cells = [np.hsplit(row, 100) for row in np.vsplit(img, 50)]
# 【2】确定trainData和testData
train_cells = [i[:50] for i in cells]
test_cells = [i[50:] for i in cells]
# 【3】对所有训练图像做抗扭斜处理
deskewed = [map(deskew, row) for row in train_cells]
# 【4】计算所有训练图像的hog
hogdata = [map(hog, row) for row in deskewed]
# 【5】训练数据的特征值、标签
trainData = np.float32(hogdata).reshape(-1, 64)
responses = np.float32(np.repeat(np.arange(10), 250)[:, np.newaxis])
# 【6】svm训练
svm = cv2.ml.SVM_create() # 创建对象
svm_params = dict(kernel_type=cv2.ml.SVM_LINEAR,
svm_type=cv2.ml.SVM_C_SVC,
C=2.67, gamma=5.383)
svm.train(trainData, cv2.ml.ROW_SAMPLE, responses, params=svm_params)
svm.save('svm_data.dat')
# 【7】用同样的方法处理测试集
deskewed = [map(deskew, row) for row in test_cells]
hogdata = [map(hog, row) for row in deskewed]
testData = np.float32(hogdata).reshape(-1, bin_n * 4)
# 【8】预测
result = svm.predict(testData)
mask = result == responses
correct = np.count_nonzero(mask)
print(correct * 100.0 / result.size)
报错
:
= np.float32 trainData (hogdata) .reshape (-1, 64)
a float or parameters should be Number String
operation result:
94%