def do_high_boost_filtering(pgm_filename, A):
high_boost_filtered = spatially_filtered_fast(pgm_filename,
high_boost_filter_kernel(A))
p1 = PGMImage(pgm_filename)
high_boost_filtered.save(f"highboost-filtered-A{A}-{p1.name}")
def averaging_filtering(pgm_filename, mask_size):
p = PGMImage(pgm_filename)
k = Kernel(mask = [[1] * mask_size] * mask_size)
p_averaging_filtered = averaging_filtered(pgm_filename, k, normalize=True)
p_averaging_filtered.save(f"averaging_filtered-{mask_size}-{p.name}")
def median_filtering(pgm_filename, mask_size):
p = PGMImage(pgm_filename)
k = Kernel(mask = [[1] * mask_size] * mask_size)
p_median_filtered = median_filtered(pgm_filename, k, normalize=True)
p_median_filtered.save(f"median_filtered-{mask_size}-{p.name}")
def test_1_identity_filter():
identity_filter = Kernel(mask=[[1]])
pgm_filename = "images/lenna.pgm"
expected = PGMImage(pgm_filename)
actual = spatially_filtered_fast(pgm_filename, identity_filter)
assert expected.pixels == actual.pixels
def do_unsharp_masking(pgm_filename: str) -> PGMImage:
p_orig = PGMImage(pgm_filename)
p_lowpass = spatially_filtered_fast(pgm_filename, smoothing_kernel)
p_lowpass.save(f"lowpass-{p_lowpass.name}")
p_highpass = p_orig - p_lowpass
p_highpass.save(f"highpass-{p_highpass.name}")
return p_highpass
def spatially_filtered_fast(pgm_filename: str,
k: Kernel,
normalize=False,
truncate=False) -> PGMImage:
p = PGMImage(pgm_filename)
def c_2d_arr_from_pyobj(pyobj: List[List[int]], arr_type, row_type):
c_2d_arr = arr_type()
for i in range(len(pyobj)):
row = row_type()
for j in range(len(pyobj[0])):
row[j] = pyobj[i][j]
c_2d_arr[i] = c_pointer_to(row)
return c_2d_arr
def c_2d_arr_empty(arr_type, row_type, rows, cols):
c_2d_arr = arr_type()
for i in range(rows):
row = row_type()
for j in range(cols):
row[j] = 0
c_2d_arr[i] = c_pointer_to(row)
return c_2d_arr
C_PGMRowT = c_char * p.cols
C_PGMImageT = PointerT(C_PGMRowT) * p.rows
C_KernelRowT = c_double * k.cols
C_KernelT = PointerT(C_KernelRowT) * k.rows
c_p = c_2d_arr_from_pyobj(p.pixels, C_PGMImageT, C_PGMRowT)
c_p_2 = c_2d_arr_empty(
PointerT(c_double * p.cols) * p.rows, c_double * p.cols, p.rows,
p.cols)
c_k = c_2d_arr_from_pyobj(k.mask, C_KernelT, C_KernelRowT)
spatialfilter = ctypes.cdll.LoadLibrary("./spatialfilter.so")
spatialfilter.apply_spatial_filter(c_p, c_p_2, p.cols, p.rows, c_k, k.rows,
k.cols)
p2 = PGMImage(pgm_filename)
for i in range(p.rows):
p2.pixels[i] = [c_p_2[i][0][j] for j in range(p2.cols)]
if normalize:
p2.normalize_intensity_values()
if truncate:
p2.truncate_intensity_values()
return p2
def smooth_image_averaging(pgm_filename):
p = PGMImage(pgm_filename)
p_average_7 = spatially_filtered(pgm_filename,
average_matrix_7,
truncate=True)
p_average_7.save(f"smoothed_averaging-7-{p.name}")
p_average_15 = spatially_filtered(pgm_filename,
average_matrix_15,
truncate=True)
p_average_15.save(f"smoothed_averaging-15-{p.name}")
def spatially_filtered(pgm_filename: str,
k: Kernel,
normalize=False,
truncate=False) -> PGMImage:
"""
Return an image with the spatial filter `k` applied to it.
:param pgm_filename file name of image to perform filter on
:param k a kernel of the spatial filter we want to apply
"""
p1, p2 = PGMImage(pgm_filename), PGMImage(pgm_filename)
for i in range(p1.rows):
new_row = []
for j in range(p1.cols):
pxl = 0
for s in range(k.rows):
for t in range(k.cols):
x, y = i - int(k.rows / 2) + s, j - int(k.cols / 2) + t
# Pad edges of the image with zeros
if x < 0 or x >= p1.rows or y < 0 or y >= p1.cols:
orig_image_x_y = 0
else:
orig_image_x_y = p1.pixels[x][y]
pxl += orig_image_x_y * k.mask[s][t]
new_row.append(pxl)
p2.pixels[i] = new_row
if normalize:
p2.normalize_intensity_values()
if truncate:
p2.truncate_intensity_values()
return p2
def do_correlation(pgm_filename, mask_filename):
pmask = PGMImage(mask_filename)
def correlation_kernel(p: PGMImage):
mask = []
for row in p.pixels:
mask.append([b for b in row])
return Kernel(mask)
k = correlation_kernel(pmask)
p2: PGMImage = spatially_filtered_fast(pgm_filename, k)
p2.save(f"correlated-to-{pmask.name}-{p2.name}")
def smooth_image_gaussian(pgm_filename):
p = PGMImage(pgm_filename)
p_gaussian = spatially_filtered(pgm_filename,
gaussian_matrix_7,
normalize=True,
truncate=False)
p_gaussian.save(f"smoothed_gaussian-7-{p.name}")
p_gaussian = spatially_filtered(pgm_filename,
gaussian_matrix_15,
normalize=True,
truncate=False)
p_gaussian.save(f"smoothed_gaussian-15-{p.name}")
def square_image(sq_sz):
canvas = np.zeros((512, 512))
for i in range(sq_sz):
for j in range(sq_sz):
x = int(256 - (sq_sz / 2) + i)
y = int(256 - (sq_sz / 2) + j)
canvas[x][y] = 255
p = PGM.PGMImage('lenna.pgm')
p.pixels = canvas
p.rows = p.cols = 512
p.truncate()
p.save(f'square-{sq_sz}.pgm')
return p
def adjust_quantization_and_save(img_name: str, quantization: int):
try:
p = PGMImage(img_name)
print(f"Adjusting quantization of {img_name} to {quantization}")
p.quantization = quantization
for i in range(p.rows):
# Normalize, quantize, and cast to byte string
p.pixels[i] = b"".join([
bytes([int((int(px) / 256) * quantization)])
for px in p.pixels[i]
])
p.save(f"images/quantized-{quantization}-{p.name}")
except IOError:
pass
def do_sharpening(pgm_filename):
p1 = PGMImage(pgm_filename)
# Prewitt filtering
p1_x_filtered_prewitt = spatially_filtered_fast(pgm_filename,
prewitt_kernel_x)
p1_x_filtered_prewitt.save(f"gradient-magnitude-prewitt-x-{p1.name}")
p1_y_filtered_prewitt = spatially_filtered_fast(pgm_filename,
prewitt_kernel_y)
p1_y_filtered_prewitt.save(f"gradient-magnitude-prewitt-y-{p1.name}")
p2 = p1_x_filtered_prewitt + p1_y_filtered_prewitt
p2.save(f"isotropic-gradient-magnitude-prewitt-{p1.name}")
p3 = p1 - p2
p3.save(f"prewitt-sharpened-{p1.name}")
# Sobel filtering
p1_x_filtered_sobel = spatially_filtered_fast(pgm_filename, sobel_kernel_x)
p1_x_filtered_sobel.save(f"gradient-magnitude-sobel-x-{p1.name}")
p1_y_filtered_sobel = spatially_filtered_fast(pgm_filename, sobel_kernel_y)
p1_y_filtered_sobel.save(f"gradient-magnitude-sobel-y-{p1.name}")
p2 = p1_x_filtered_sobel + p1_y_filtered_sobel
p2.save(f"isotropic-gradient-magnitude-sobel-{p1.name}")
p3 = p1 - p2
p3.save(f"sobel-sharpened-{p1.name}")
# Laplace filtering
p1_laplace_filtered = spatially_filtered_fast(pgm_filename,
laplacian_kernel)
p1_laplace_filtered.save(f"laplacian-gradient-magnitude-{p1.name}")
p2 = p1 - p1_laplace_filtered
p2.save(f"laplacian-sharpened-{p1.name}")
def transformer_of(p: PGMImage) -> List[int]:
""" Calculate a transformer for histogram equalization.
The transformer is a list, indexed by intensity values, whose
values represent the output intensity required for an equalized
histogram, given an input histogram.
:param p image to equalize the histogram of
"""
histogram = p.get_histogram(normed=True)
L = p.quantization # Number of distinct grey levels
# Calculate histogram transformer
T_r = [0] * (L + 1) # len([0, L]) = L + 1
for i in range(len(histogram)):
if i == 0:
T_r[i] = L * histogram[i]
else:
T_r[i] = ((T_r[i - 1] / L) + histogram[i]) * L
# Discretize T_r by taking the ceil
return [min(int(t_r + 1), 255) for t_r in T_r]
def equalize_histogram(img_name: str, visualize_results: bool):
p = PGMImage(img_name)
T_r = transformer_of(p)
p2 = PGMImage(img_name)
# Transform the image according to T_r
for i in range(len(p2.pixels)):
p2.pixels[i] = b"".join(bytes([T_r[b]]) for b in p2.pixels[i])
p2.save(f"equalized-{p2.name}")
if visualize_results:
p.show_histogram(title=f"Histogram of {p.name} before equalization")
p2.show_histogram(title=f"Histogram of {p2.name} after equalization")
def specify_histogram(img_name, specified_img_name, visualize_results=True):
p1, p2 = PGMImage(img_name), PGMImage(specified_img_name)
T_r = transformer_of(p1)
G_z = transformer_of(p2)
def inverted_histogram(h: List[int]) -> List[int]:
G_i = [0] * len(h)
for h_i in h:
G_i[h_i] = h_i
# Populate missing values
last_nonzero_intensity = 0
for i in range(len(h)):
if G_i[i] != 0:
last_nonzero_intensity = G_i[i]
else:
G_i[i] = last_nonzero_intensity
return G_i
G_inverse_z = inverted_histogram(G_z)
p3 = PGMImage(img_name)
for i in range(len(p3.pixels)):
p3.pixels[i] = b"".join(
bytes([G_inverse_z[T_r[b]]]) for b in p3.pixels[i])
p3.save(f"specified-to-{p2.name}-{p1.name}")
if visualize_results:
p1.show_histogram()
p2.show_histogram()
p3.show_histogram(
title=f"Histogram of {p1.name}, specified to {p2.name}")
p = PGM('zigzag.pgm')
a = np.fft.fft2(p.pixels)
ashift = np.fft.fftshift(a)
magnitude_spectrum = 20 * np.log(np.abs(ashift))
# Set Plot properties
# plt.scatter(a.real, a.imag)
# plt.axis([-1_000_000, 1_000_000, -400_000, 400_000])
plt.imshow(20 * np.log(np.abs(np.fft.fftshift(a))), cmap='gray')
# Display plot
plt.show()
from PGM import PGMImage
p = PGMImage('lenna.pgm')
pxls = p.pixels
import copy
fft_out_c = copy.deepcopy(pxls)
fft_out_py = np.fft.fft2(pxls)
fft_out_c = [my_cfft(row) for row in fft_out_c]
for i in range(p.cols):
col = [row[i] for row in fft_out_c]
col = my_cfft(col)
for j in range(p.rows):
fft_out_c[j][i] = col[j]
