def affine_transform(img):
#print profile_path, name
rows, cols, ch = img.shape
x_rot = 50
y_rot = 200
pts1 = np.float32([[50, 50], [200, 50], [50, 200]])
x_rot = x_rot + 20
y_rot = y_rot + 20
pts2 = np.float32([[x_rot, x_rot], [200, 50], [x_rot, y_rot]])
M = cv2.getAffineTransform(pts1, pts2)
dst = cv2.warpAffine(img, M, (cols, rows))
salt = saltandpepper1.noise_addition(dst)
#cv2.imshow('salt', salt)
#cv2.waitKey()
con_img = convolve_image(salt)
#cv2.imshow('consalt', con_img)
#cv2.waitKey()
return con_img
def generate_affine(img, name):
#for i in range(10):
for i in range(30):
#padd the image
img1 = padd(img)
#cv2.imshow('after concatat',img1)
#cv2.waitKey()
gamma_1 = random_1()
gamma_main = random.randrange(1, 3, 1)
gamma_last = gamma_main + gamma_1
noise_addition = adjust_gamma(
img1, gamma=gamma_last) #change the gammmaa of the image randomly
noise_ad2 = sp.noise_addition(
noise_addition) #add the salt and peppr noise to the image
pts1 = np.float32([[50, 50], [200, 50], [50, 200]])
(h, w) = img1.shape[:2]
rows2, cols2, ch2 = img1.shape #ch=3 rows = hieght and col is width
#creating the rotation matrix
x = random.randrange(-3, 3, 1)
y = random.random()
z = x + y
x = math.radians(z)
sin = math.sin(x)
cos = math.cos(x)
rotation = [[cos, sin], [-sin, cos]] #2x2 rotation matrix
#shear matrices
shear_fact_x = -0.05 + random_2() #take random values for both x and y
shear_fact_y = -0.05 + random_2()
shear_x = [[1, shear_fact_x], [0, 1]]
shear_y = [[1, 0], [shear_fact_y, 1]]
shear_new = np.array(shear_x, dtype=np.float32)
rotation_new = np.array(rotation, dtype=np.float32)
#multiplying 2D matrices
afine_t = np.dot(shear_new, rotation_new)
shear_new = np.array(shear_y, dtype=np.float32)
afine_t2 = np.dot(afine_t, shear_new)
#afine_t = shear_new
pts2 = np.dot(pts1, afine_t2) #new points use for warping the image
#print pts2
#cv2.imshow('Affine_2',img)
#cv2.waitKey()
s11, s12, s13 = noise_ad2.shape
for r in range(s11):
for c in range(s12):
if np.all(noise_ad2[r, c, :] == 0):
noise_ad2[r, c, 0] = noise_ad2[r, c, 0] + 1
noise_ad2[r, c, 1] = noise_ad2[r, c, 1] + 1
noise_ad2[r, c, 2] = noise_ad2[r, c, 2] + 1
M = cv2.getAffineTransform(pts1, pts2)
dst = cv2.warpAffine(noise_ad2, M, (cols2, rows2)) #get afine of image
#cv2.imshow('imshow',dst)
#cv2.imshow('Affine_3',dst)
#cv2.waitKey()
s1, s2, s3 = dst.shape
#removing the extra black areas from the images so that only logos is visible.
count = 0
count2 = 0
for r in range(s1):
num = count_rows(dst[r, :, :])
#print '*******NUM*******', num
if num >= 30:
count = count + 1
#break
else:
#count = count+1
break
for c in range(s2):
num2 = count_rows(dst[:, c, :])
#print '*******NUM*******1\t', num2
if (num2 >= 30):
count2 = count2 + 1
#break
else:
#count2 = count2+1
break
r1 = s1 - 1
while (r1 >= 0):
num3 = count_rows(dst[r1, :, :])
#print '*******NUM*******2\t', num3
if (num3 >= 30):
#break
r1 = r1 - 1
else:
#r1=r1-1
break
c1 = s2 - 1
while (c1 >= 0):
num4 = count_rows(dst[:, c1, :])
#print '*******NUM*******3\t', num4
if (num4 >= 30):
#break
c1 = c1 - 1
else:
#c1 = c1-1
break
# print 'first and last also col1 and col2',r1,count,count2,c1
dst2 = dst[count:r1, count2:c1]
#major change
or_image_x, or_image_y, or_image_z = img.shape
#dst2 = dst
s1, s2, s3 = dst2.shape
t2 = (-or_image_y + s2) / 2
t1 = (-or_image_x + s1) / 2
#print '***t2,t2***', t1, t2
count = r1 = c1 = count2 = 0
#print '*******s1*******', s1, s2, s3
if s1 > 100 and s2 > 100:
dst2 = dst2[t1:s1 - t1, t2:s2 - t2]
#cv2.imshow('Affine',dst2)
#cv2.waitKey()
else:
#print '******AFFINE FAILED********'
dst2 = img #comping original image to dst
#add the gist and hog feature of the image
if s2 != 0 and s1 != 0:
# cv2.imshow('noise',dst2)
#dst2 remove black.
gist_feat = Gist_feat_last.singleImage2(dst2)
hog_feat = HOG_feat2.hog_call(dst2)
gist_feat = np.concatenate((gist_feat, [1]))
hog_feat = np.concatenate((hog_feat, [1]))
'''add_feature_to_database(name,gist_feat,"Gabor")
add_feature_to_database(name,hog_feat,"HOG")'''
with open(feat_gist + name + '.csv', 'a') as myfile:
wr = csv.writer(myfile, quoting=csv.QUOTE_ALL)
wr.writerow(gist_feat)
with open(feat_hog + name + '.csv', 'a') as myfile:
wr = csv.writer(myfile, quoting=csv.QUOTE_ALL)
wr.writerow(hog_feat)
return i