def ideal_filter(fp, param, dst=None):
if not os.path.exists(fp):
raise Exception('[Error] image file not exists')
img = cv2.cvtColor(cv2.imread(fp, cv2.IMREAD_COLOR), cv2.COLOR_BGR2GRAY)
w, h = img.shape
img_dft = np.fft.fft2(center_transform(img))
result = np.copy(img_dft)
if param.mode == 'lo':
for i in range(w):
for j in range(h):
if math.sqrt((i - w / 2)**2 +
(j - h / 2)**2) > param.threshold:
result[i, j] = 0 + 0.j
elif param.mode == 'hi':
for i in range(w):
for j in range(h):
if math.sqrt((i - w / 2)**2 +
(j - h / 2)**2) <= param.threshold:
result[i, j] = 0 + 0.j
r_comp = np.real(np.fft.ifft2(result))
img_ct = center_transform(r_comp, dst)
return img_ct
def butterworth_filter(fp, param, dst=None):
if not os.path.exists(fp):
raise Exception('[Error] image file not exists')
img = cv2.cvtColor(cv2.imread(fp, cv2.IMREAD_COLOR), cv2.COLOR_BGR2GRAY)
w, h = img.shape
img_dft = np.fft.fft2(center_transform(img))
result = np.copy(img_dft)
if param.mode == 'lo':
for i in range(w):
for j in range(h):
tmp = (i - w / 2) ** 2 + (j - h / 2) ** 2
result[i, j] = img_dft[i, j] * 1 / (1 + (math.sqrt(tmp) / param.threshold) ** (2 * param.dim))
elif param.mode == 'hi':
for i in range(w):
for j in range(h):
tmp = (i - w / 2) ** 2 + (j - h / 2) ** 2
if tmp == 0:
result[i, j] = 0 + 0.j
continue
result[i, j] = img_dft[i, j] * 1 / (1 + (param.threshold / math.sqrt(tmp)) ** (2 * param.dim))
r_comp = np.real(np.fft.ifft2(result))
img_ct = center_transform(r_comp, dst)
return img_ct
def inverse_filter(img, hm, dst=None):
img_dft = np.fft.fft2(center_transform(img))
fm = img_dft / hm
r_comp = np.real(np.fft.ifft2(fm))
img_ct = center_transform(r_comp, dst)
return img_ct
def wiener_deconv_filter(img, fm, hm, noise, dst=None):
img_dft = np.fft.fft2(center_transform(img))
sf = abs(np.fft.fft2(center_transform(fm)))**2
sn = abs(np.fft.fft2(center_transform(noise)))**2
result = ((abs(hm)**2) / hm / (abs(hm)**2 + sn / sf)) * img_dft
r_comp = np.real(np.fft.ifft2(result))
img_ct = center_transform(r_comp, dst)
return img_ct
def blurring_filter(img, a=0.1, b=0.1, t=1, dst=None):
eps = 0.000001
w, h = img.shape
img_dft = np.fft.fft2(center_transform(img))
hm = np.copy(img_dft)
gm = np.copy(img_dft)
for i in range(w):
for j in range(h):
theta = math.pi * (a * (i - w / 2) + b * (j - h / 2))
hm[i, j] = 1 + 0.j if abs(theta) < eps else (
t / theta) * math.sin(theta) * cmath.exp(-1.j * theta)
gm[i, j] = img_dft[i, j] * hm[i, j]
r_comp = np.real(np.fft.ifft2(gm))
img_ct = center_transform(r_comp, dst)
return img_ct, hm