def HOG_draw(gr, img):
gray = lib0110.BGR2GRAY(img)
add = np.zeros_like(gray)
C, height, width = gr.shape
for y in range(height):
for x in range(width):
mat = np.zeros((8,8))
for c in range(C):
score = gr[c,y,x] * 10
angle = np.pi * (20*c+10) /180
print(angle)
for xx in range(-4,4):
yy = xx * np.tan(angle)
yy = int(np.round(yy))
if yy>=-4 and yy<4:
mat[yy+4, xx+4] = max(score, mat[yy+4, xx+4])
for yy in range(-4,4):
a = np.tan(angle)
if a==0:
a = 1e-5
xx = yy / a
xx = int(np.round(xx))
if xx>=-4 and xx<4:
mat[yy+4, xx+4] = max(score, mat[yy+4, xx+4])
add[y*8:(y+1)*8, x*8:(x+1)*8] = mat
return add
def Harris_corner_detection2(img, ix2g, iy2g, ixyg, k, threshold):
gray = lib0110.BGR2GRAY(img)
height, width = gray.shape
out = gray.copy().ravel()
out = out.repeat(3).reshape(height, width, 3)
R = (ix2g * iy2g - ixyg * ixyg) - k * ((ix2g + iy2g)**2)
out[R >= np.max(R) * threshold] = (0, 0, 255)
return out
def Gabor_filtering(img, K, s, g, l, p, A):
gray = lib0110.BGR2GRAY(img)
filter = make_Gabor_filter(K,s,g,l,p,A)
hK = K//2
out = np.zeros_like(gray)
height, width = gray.shape
pad = np.pad(gray, [(hK,hK), (K-hK,K-hK)], 'edge')
for y in range(height):
for x in range(width):
out[y,x] = np.sum(pad[y:y+K, x:x+K] * filter)
return out
def Saliency_map(img):
gray = lib0110.BGR2GRAY(img)
pyramids = []
pyramids.append(gray.clip(0,255).astype(np.uint8))
for i in range(1,6):
out2 = lib2130.Bi_linear_interpolation(gray, 1./(2**i), 1./(2**i))
out3 = lib2130.Bi_linear_interpolation(out2, (2**i), (2**i))
pyramids.append(out3)
dif = np.zeros_like(gray)
for i in range(5):
for j in range(i,6):
dif += np.abs(pyramids[j] - pyramids[i])
return lib2130.Histogram_normalization(dif, 0, 255)
def Canny_edge_strength(img, k, sigma):
img = img.astype(np.float64)
img1 = lib0110.BGR2GRAY(img)
img2 = lib0110.Gaussian_filter(img1, k=k, sigma=sigma, padding_type='edge')
img3v, img3h = lib1120.Sobel_filter(img2, padding_type='edge')
edge = np.sqrt(img3h * img3h + img3v * img3v)
img3h[img3h == 0] = 1e-5
tan = np.arctan(img3v / img3h)
angle = np.zeros_like(tan)
angle[np.where((-0.4142 < tan) & (tan <= 0.4142))] = 0
angle[np.where((0.4142 < tan) & (tan < 2.4142))] = 45
angle[np.where((tan <= -2.4142) | (2.4142 <= tan))] = 90
angle[np.where((-2.4142 < tan) & (tan <= -0.4142))] = 135
return edge, angle
def Harris_corner_detection1(img, Gk, Gsigma):
gray = lib0110.BGR2GRAY(img)
height, width = gray.shape
iy, ix = lib1120.Sobel_filter(gray, padding_type='edge')
ix2g = lib0110.Gaussian_filter(ix * ix,
k=Gk,
sigma=Gsigma,
padding_type='edge')
iy2g = lib0110.Gaussian_filter(iy * iy,
k=Gk,
sigma=Gsigma,
padding_type='edge')
ixyg = lib0110.Gaussian_filter(ix * iy,
k=Gk,
sigma=Gsigma,
padding_type='edge')
return ix2g, iy2g, ixyg
def Hessian_corner_detection(img, threshold):
gray = lib0110.BGR2GRAY(img)
height, width = gray.shape
iy, ix = lib1120.Sobel_filter(gray, padding_type='edge')
iyy, ixy = lib1120.Sobel_filter(iy, padding_type='edge')
_, ixx = lib1120.Sobel_filter(ix, padding_type='edge')
detH = iyy * ixx - ixy * ixy
maxdH = np.max(detH)
out = gray.copy().ravel()
out = out.repeat(3).reshape(height, width, 3)
for y in range(height):
for x in range(width):
neighbor = detH[max(0, y - 1):min(y + 2, height),
max(0, x - 1):min(x + 2, width)]
if np.max(neighbor) == detH[y,
x] and detH[y, x] >= threshold * maxdH:
out[y, x] = (0, 0, 255)
return out
def HOG1_gradient(img):
gray = lib0110.BGR2GRAY(img).astype(np.float32)
height, width = gray.shape
pad = np.pad(gray, [(1,1),(1,1)], 'edge')
gx = pad[1:-1, 2:] - pad[1:-1, :-2]
gy = pad[2:, 1:-1] - pad[:-2, 1:-1]
mag = np.sqrt(gx*gx + gy*gy)
gx[gx==0] = 1e-6
ang = np.arctan(gy / gx)
ang[ang < 0] = np.pi + ang[ang < 0]
a2 = np.zeros_like(ang)
for i in range(9):
index = np.where((ang > (np.pi/9)*i) & (ang <=(np.pi/9)*(i+1)))
a2[index] = i
return mag, a2
import cv2
import numpy as np
import lib5160
import matplotlib.pyplot as plt
import sys,os
sys.path.append(os.getcwd())
from q01_10 import lib0110
from q41_50 import lib4150
if __name__ == '__main__':
img = cv2.imread("Gasyori100knock/Question_51_60/imori.jpg")
img2 = lib0110.OTSU_binarization(lib0110.BGR2GRAY(img))
out = lib5160.Morphology_gradient(img2).clip(0,255).astype(np.uint8)
cv2.imshow("imori", out)
cv2.waitKey(0)
cv2.imwrite("q51_60/051.jpg", out)
cv2.destroyAllWindows()
import cv2
import numpy as np
import lib1120
import sys,os
sys.path.append(os.getcwd())
from q01_10 import lib0110
if __name__ == '__main__':
img = cv2.imread("Gasyori100knock/Question_11_20/imori.jpg")
gray_img = lib0110.BGR2GRAY(img)
ansv, ansh = lib1120.Sobel_filter(gray_img)
ansv = ansv.clip(0,255).astype(np.uint8)
ansh = ansh.clip(0,255).astype(np.uint8)
cv2.imshow("imori", ansv)
cv2.waitKey(0)
cv2.imwrite("q11_20/015v.jpg", ansv)
cv2.imshow("imori", ansh)
cv2.waitKey(0)
cv2.imwrite("q11_20/015h.jpg", ansh)
cv2.destroyAllWindows()
import cv2
import numpy as np
import lib7180
import matplotlib.pyplot as plt
import sys,os
sys.path.append(os.getcwd())
from q21_30 import lib2130
from q01_10 import lib0110
from q61_70 import lib6170
if __name__ == '__main__':
img = cv2.imread("Gasyori100knock/Question_71_80/imori.jpg")
gray = lib0110.BGR2GRAY(img)
pyramids = []
pyramids.append(gray.clip(0,255).astype(np.uint8))
for i in range(1,6):
out2 = lib2130.Bi_linear_interpolation(gray, 1./(2**i), 1./(2**i)).clip(0,255).astype(np.uint8)
pyramids.append(out2)
for i in range(6):
cv2.imshow("imori", pyramids[i])
cv2.waitKey(0)
cv2.imwrite("q71_80/075_"+str(i)+".jpg", pyramids[i])
cv2.destroyAllWindows()
import cv2
import numpy as np
import lib6170
import matplotlib.pyplot as plt
import sys, os
sys.path.append(os.getcwd())
from q01_10 import lib0110
from q51_60 import lib5160
if __name__ == '__main__':
img = cv2.imread("Gasyori100knock/Question_61_70/imori.jpg")
out = lib6170.HOG(img).clip(0, 255).astype(np.uint8)
out = lib5160.Alpha_blend(lib0110.BGR2GRAY(img), out, 0.2)
cv2.imshow("imori", out)
cv2.waitKey(0)
cv2.imwrite("q61_70/069.jpg", out)
cv2.destroyAllWindows()
def Pyramid_difference(img, a):
gray = lib0110.BGR2GRAY(img)
zoi = zoomout_in(img, a)
out = np.abs(zoi - gray)
return lib2130.Histogram_normalization(out, 0, 255)
def zoomout_in(img, a):
gray = lib0110.BGR2GRAY(img)
zoomout = lib2130.Bi_linear_interpolation(gray, 1./a, 1./a)
zoomin = lib2130.Bi_linear_interpolation(zoomout, a, a)
return zoomin
