class SVM(Symbol):
ratio = 0.8
def __init__(self):
self.model = LinearSVC()
def load(self, fn):
r = np.load(fn)
classes_ = r['classes']
coef_ = r['coef']
intercept_ = r['intercept']
self.model.__setattr__('classes_', classes_)
self.model.__setattr__('coef_', coef_)
self.model.__setattr__('intercept_', intercept_)
def train(self, samples, responses):
self.model.fit(samples, responses)
def predict(self, samples):
return self.model.predict(samples)
def save(self, path):
classes_ = self.model.classes_
coef_ = self.model.coef_
intercept_ = self.model.intercept_
np.savez(path, classes=classes_, coef=coef_, intercept=intercept_)
class SymbolRecognition(object):
def __init__(self):
# init the cv2 hog class
self.winSize = (64, 64)
blockSize = (16, 16)
blockStride = (8, 8)
cellSize = (8, 8)
nbins = 9
self.hog = cv2.HOGDescriptor(self.winSize, blockSize, blockStride,
cellSize, nbins)
# init the sklearn linearSVC
modelpath = 'model.npz'
self.model = LinearSVC()
self.load_model(modelpath)
def load_model(self, path):
r = np.load(path)
classes_ = r['classes']
coef_ = r['coef']
intercept_ = r['intercept']
self.model.__setattr__('classes_', classes_)
self.model.__setattr__('coef_', coef_)
self.model.__setattr__('intercept_', intercept_)
def extract_hog(self, im):
descriptor = self.hog.compute(im)
if descriptor is None:
return None
return descriptor.ravel()
def predict(self, im):
im = cv2.resize(im, self.winSize, interpolation=cv2.INTER_CUBIC)
feature = self.extract_hog(im)
return self.model.predict(feature.reshape(1, -1))[0]
