2장 OpenCV를 이용한 기하학적 변환
import numpy as np import cv2 import sys import matplotlib.pyplot as plt from pathlib import Path 영상 필터링 (Filtering)
2D Convolution
# cv2.filter2D(src, ddepth, kernel[, dst[, anchor[, delta[, borderType]]]]) -> dst # src: 입력영상 # ddepth: 출력영상의 타입 (cv2.CV_8U, cv2.CV_32F, cv2.CV_64F), # -1 => 입력영상과 같은 타입 # kernel: filter 행렬, 실수형 # anchor: (-1, -1) 필터의 중앙점 # delta: 더할 값 # borderType: 가장자리 픽셀확장 방식 # dst: 출력영상
# blur(src, ksize[, dst[, anchor[, borderType]]]) -> dst # src: 입력영상 # ksize: mean filter kernel size folder = "fig" ## 영역처리, 영상 필터링, Convolution
# src = cv2.imread("./fig/blue_eyes.png", cv2.IMREAD_REDUCED_COLOR_2) src = cv2.imread(Path(folder, "blue_eyes.png" ), cv2. IMREAD_REDUCED_COLOR_2 )
kernel_3 = np.ones(( 3 , 3 ), dtype = np.float64) / 9 . dst_kernel = cv2.filter2D(src, - 1 , kernel_3)
dst_blur = cv2.blur(src, ( 3 , 3 ))
cv2.imshow( 'src' , src) cv2.imshow( 'dst_kernel' , dst_kernel) cv2.imshow( "dst_blur" , dst_blur)
cv2.waitKey() cv2.destroyAllWindows() src = cv2.imread(Path(folder, "blue_eyes.png" ), cv2. IMREAD_REDUCED_COLOR_2 ) cv2.imshow( "src" , src)
for ksize in ( 3 , 5 , 7 , 9 , 12 , 15 , 23 ): dst = cv2.blur(src, (ksize, ksize)) text = ' {} x {} ' .format(ksize, ksize) cv2.putText(dst, text, ( 20 , 30 ), cv2. FONT_HERSHEY_COMPLEX , 1 , ( 0 , 0 , 255 ), 2 , cv2. LINE_AA )
cv2.imshow( 'mean filter' , dst) cv2.waitKey()
cv2.destroyAllWindows() 가우시안 필터
# GaussianBlur(src, (ksize), sigmaX[, dst[, sigmaY[, borderType]]]) -> dst # src: 입력영상 # ksize: mean filter kernel size, (0, 0) 자동으로 결정 # sigmaX: gaussian x 방향의 sigma # sigmaY: gaussian y 방향의 sigma ## 가우시안 필터
src = cv2.imread(Path(folder, "blue_eyes.png" ), cv2. IMREAD_COLOR )
dst_Gaussian1 = cv2.GaussianBlur(src, ( 0 , 0 ), 1 ) dst_mean = cv2.blur(src, ( 5 , 5 )) # dst_Gaussian2 = cv2.GaussianBlur(src, (0, 0), 2) # dst_Gaussian3 = cv2.GaussianBlur(src, (0, 0), 3)
cv2.imshow( 'src' , src) cv2.imshow( "Gaussian1" , dst_Gaussian1) cv2.imshow( "dst_mean" , dst_mean) # cv2.imshow("Gaussian2", dst_Gaussian2) # cv2.imshow("Gaussian3", dst_Gaussian3)
cv2.waitKey() cv2.destroyAllWindows() 사프닝 필터링
## Sharpening filter
src = cv2.imread(Path(folder, "blue_eyes.png" ), cv2. IMREAD_REDUCED_COLOR_2 )
blr = cv2.GaussianBlur(src, ( 0 , 0 ), 1 ) blr2 = cv2.GaussianBlur(src, ( 0 , 0 ), 2 ) blr3 = cv2.GaussianBlur(src, ( 0 , 0 ), 3 )
dst = cv2.addWeighted(src, 2 , blr, - 1 , 0.0 ) dst2 = cv2.addWeighted(src, 2 , blr2, - 1 , 0.0 ) dst3 = cv2.addWeighted(src, 2 , blr3, - 1 , 0.0 )
cv2.imshow( 'src' , src) cv2.imshow( "blr" , blr) cv2.imshow( "sharpening1" , dst) cv2.imshow( "sharpening2" , dst2) cv2.imshow( "sharpening3" , dst3)
cv2.waitKey() cv2.destroyAllWindows()
# [[ 0. -1. 0.] # [-1. 5. -1.] # [ 0. -1. 0.]]
# [[-0.2 -0.8 -0.2] # [-0.8 5. -0.8] # [-0.2 -0.8 -0.2]]
# src = cv2.imread('blue_eyes.png', cv2.IMREAD_REDUCED_COLOR_2) # kernel = np.ones((3, 3), dtype = np.float64)*-1 # kernel[1,1] = 9 # #print(kernel) # dst = cv2.filter2D(src, -1, kernel) # cv2_imshow(src) # cv2_imshow(dst) ## sharpen filter src = cv2.imread(Path(folder, "blue_eyes.png" ))
kernel_3 = np.ones(( 3 , 3 ), dtype = np.float64) *- .5 kernel_3[ 1 , 1 ] = 5
dst_kernel = cv2.filter2D(src, - 1 , kernel_3)
cv2.imshow( 'src' , src) cv2.imshow( 'dst_kernel' , dst_kernel)
plt.imshow(kernel_3, cmap = "seismic" ) plt.colorbar() plt.show() # cv2.imshow("dst_blur", dst_blur)
cv2.waitKey() cv2.destroyAllWindows() print (kernel_3)
[[-0.5 -0.5 -0.5]
[-0.5 5. -0.5]
[-0.5 -0.5 -0.5]]
중앙값 필터링
## median filter src = cv2.imread(Path(folder, "blue_eyes.png" ), cv2. IMREAD_GRAYSCALE ) salt_noise = np.random.choice(( 0 , 255 ), src.shape, p = ( 0.99 , 0.01 )).astype(np.uint8) pepper_noise = np.random.choice(( 0 , 255 ), src.shape, p = ( 0.99 , 0.01 )).astype(np.uint8) dst = cv2.add(src, salt_noise) dst = cv2.subtract(dst, pepper_noise)
dst_median = cv2.medianBlur(dst, 3 ) # dst_Gaussian = cv2.GaussianBlur(dst, (0, 0), 1)
cv2.imshow( "src" , src) # cv2.imshow("salt_noise", salt_noise) # cv2.imshow("pepper_noise", pepper_noise) cv2.imshow( "dst" , dst) cv2.imshow( "dst_median" , dst_median) # cv2.imshow("dst_Gaussian", dst_Gaussian)
while True : if cv2.waitKey() == ord ( "q" ): break cv2.destroyAllWindows() 양방향 필터 (Bilateral filter)
# bilateralFilter(src, d, sigmaColor, sigmaSpace[, dst[, borderType]]) -> dst # src: 입력영상 # d: -1로 설정 # sigmaColor: 색공간의 표준편차 # sigmaSpace: 좌표공간의표준편차 # dst: # borderType: 가장자리 픽셀확장 방식 src = cv2.imread(Path(folder, "blue_eyes.png" ), cv2. IMREAD_GRAYSCALE )
dst_Gaussian = cv2.GaussianBlur(src, ( 0 , 0 ), 1 . ) dst_Bilateral_5 = cv2.bilateralFilter(src, - 1 , 5 , 10 ) dst_Bilateral_20 = cv2.bilateralFilter(src, - 1 , 20 , 10 )
cv2.imshow( "src" , src) cv2.imshow( "Gaussian" , dst_Gaussian) cv2.imshow( "Bilateral_5" , dst_Bilateral_5) cv2.imshow( "Bilateral_20" , dst_Bilateral_20)
cv2.waitKey() cv2.destroyAllWindows() 카툰 필터
## bitwise operator src1 = np.zeros(( 256 , 256 ), np.uint8) cv2.rectangle(src1, ( 10 , 10 ), ( 127 , 245 ), 255 , - 1 ) src2 = np.zeros(( 256 , 256 ), np.uint8) cv2.circle(src2, ( 127 , 127 ), 100 , 128 , - 1 )
dst1 = cv2.bitwise_and(src1, src2) dst2 = cv2.bitwise_or(src1, src2) dst3 = cv2.bitwise_xor(src1, src2) dst4 = cv2.bitwise_not(src2)
cv2.imshow( "src1" , src1) cv2.imshow( 'src2' , src2) cv2.imshow( "AND" , dst1) cv2.imshow( "OR" , dst2) cv2.imshow( "XOR" , dst3) cv2.imshow( "NOT" , dst4)
cv2.waitKey() cv2.destroyAllWindows() ## Cartoon 필터
time = cv2.TickMeter()
# src = cv2.imread("./fig/son.jpg") src = cv2.imread(Path(folder, "son.jpg" ))
time.start()
### 연산 구간 for i in range ( 100 ): blr = cv2.GaussianBlur(src, ( 0 , 0 ), 2 ) edge = 255 - cv2.Canny(src, 100 , 200 ) edge = cv2.cvtColor(edge, cv2. COLOR_GRAY2BGR ) dst = cv2.bitwise_and(blr, edge) ###
time.stop()
cv2.imshow( "son" , src) cv2.imshow( "blr" , blr) cv2.imshow( "edge" , edge) cv2.imshow( "dst" , dst) print (time.getTimeMilli(), "ms" ) time.reset()
cv2.waitKey() cv2.destroyAllWindows() 260.1467 ms
영상 이동 (Translation)
# warpAffine(src, M, dsize[, dst[, flags[, borderMode[, borderValue]]]]) -> dst # src: 입력영상 # M: affine transform matrix (size: 2 x 3) # dsize: 출력영상 크기, (0, 0) = 입력영상크기로 출력 # borderValue: 값이 없는 영역을 채우는 값, default = 0 ## 영상 이동 (Translation)
# src = cv2.imread("./fig/dog.bmp") src = cv2.imread(Path(folder, "dog.bmp" ))
## affine matrix affine = np.array([[ 1 , 0 , 50 ], [ 0 , 1 , 200 ]], np.float32)
print ( "affine matrix: \n " , affine) # print(affine.dtype)
dst = cv2.warpAffine(src, affine, ( 0 , 0 ), borderMode = cv2. BORDER_CONSTANT , borderValue = ( 255 , 255 , 255 ))
# dst = cv2.warpAffine(src, affine, (0, 0))
cv2.imshow( "src" , src) cv2.imshow( "dst" , dst)
cv2.waitKey() cv2.destroyAllWindows() affine matrix:
[[ 1. 0. 50.]
[ 0. 1. 200.]]
영상 회전 (Rotation)
# warpAffine(src, M, dsize[, dst[, flags[, borderMode[, borderValue]]]]) -> dst # src: 입력영상 # M: affine transform matrix (size: 2 x 3) # dsize: 출력영상 크기, (0, 0) = 입력영상크기로 출력 # borderValue: 값이 없는 영역을 채우는 값, default = 0
# getRotationMatrix2D(center, angle, scale) -> retval # center: 영상의 center # angle: 회전각도 # scale: 확대율 ## 회전 변환 (Rotation)
src = cv2.imread(Path(folder, "dog.bmp" ))
cp = (src.shape[ 1 ] / 2 , src.shape[ 0 ] / 2 ) affine = cv2.getRotationMatrix2D(cp, 30 , 1 .) dst = cv2.warpAffine(src, affine, ( 0 , 0 ))
cv2.imshow( "src" , src) cv2.imshow( "dst" , dst)
cv2.waitKey() cv2.destroyAllWindows() 영상 확대 (Scaling)
## 영상 확대 (Scaling) # src = cv2.imread("./fig/rose.jpg") src = cv2.imread(Path(folder, "rose.jpg" ))
print ( "src shape : " , src.shape)
dst1 = cv2.resize(src, ( 1900 , 1300 ), interpolation = cv2. INTER_NEAREST ) dst2 = cv2.resize(src, ( 1900 , 1300 ), interpolation = cv2. INTER_LINEAR ) dst3 = cv2.resize(src, ( 1900 , 1300 ), interpolation = cv2. INTER_CUBIC )
cv2.imshow( "src" , src) cv2.imshow( 'INTER_NEAREST' , dst1[ 800 : 1600 , 800 : 1200 ]) cv2.imshow( 'INTER_LINEAR' , dst2[ 800 : 1600 , 800 : 1200 ]) cv2.imshow( 'INTER_CUBIC' , dst3[ 800 : 1600 , 800 : 1200 ])
cv2.waitKey() cv2.destroyAllWindows() ---------------------------------------------------------------------------
NameError Traceback (most recent call last)
Cell In[3], line 3
1 ## 영상 확대 (Scaling)
2 # src = cv2.imread("./fig/rose.jpg")
----> 3 src = cv2.imread(Path(folder, "rose.jpg"))
5 print("src shape : ", src.shape)
7 dst1 = cv2.resize(src, (1900, 1300), interpolation=cv2.INTER_NEAREST)
NameError: name 'folder' is not defined
영상 전단 (Shearing)
## Shearing src = cv2.imread(Path(folder, "dog.bmp" ))
affine = np.array([[ 1 , 0.2 , 0 ], [ 0.2 , 1 , 0 ]], np.float32)
dst = cv2.warpAffine(src, affine, ( 0 , 0 ))
cv2.imshow( "src" , src) cv2.imshow( "dst" , dst)
cv2.waitKey() cv2.destroyAllWindows() # getPerspectiveTransform(src, dst[, solveMethod]) -> retval # src: 입력영상의 4개 좌표점, numpy array shape(4,2) # dst: 출력영상의 4개 좌표점, numpy array shape(4,2) ## Non linear warping
# src = cv2.imread("./fig/checkerboard.png") src = cv2.imread(Path(folder, "checkerboard.png" ))
h, w = src.shape[: 2 ]
srcPoint = np.array([[ 218 , 48 ], [ 691 , 47 ], [ 830 , 518 ], [ 67 , 527 ]], np.float32) dstPoint = np.array([[ 0 , 0 ], [w - 1 , 0 ], [w - 1 , h - 1 ], [ 0 , h - 1 ]], np.float32)
pers = cv2.getPerspectiveTransform(srcPoint, dstPoint) # array([[ 2.13264257e+00, 6.72294421e-01, -4.97186212e+02], # [ 4.08735101e-03, 1.93331703e+00, -9.36902599e+01], # [-1.86504523e-05, 1.36282733e-03, 1.00000000e+00]])
dst = cv2.warpPerspective(src, pers, (w, h)) # a, b, c, d = cv2.selectROI(src) # print(a, b, c, d)
print ( "width = " , w, "height = " , h) cv2.imshow( "src" , src) cv2.imshow( "dst" , dst)
cv2.waitKey() cv2.destroyAllWindows() width = 960 height = 540
비선형 리매핑
# remap(src, map1, map2, interpolation[, dst[, borderMode[, borderValue]]]) -> dst ## remapping # src = cv2.imread("./fig/bamboo.jpg") src = cv2.imread(Path(folder, "bamboo.jpg" ))
h, w = src.shape[: 2 ]
map2, map1 = np.indices((h, w), np.float32) # print(map1) print ( "" ) print (map2)
map1 = map1 + 10 * np.sin(map2 / 32 )
dst = cv2.remap(src, map1, map2, cv2. INTER_CUBIC )
cv2.imshow( 'src' , src) cv2.imshow( 'dst' , dst)
cv2.waitKey() cv2.destroyAllWindows() [[ 0. 0. 0. ... 0. 0. 0.]
[ 1. 1. 1. ... 1. 1. 1.]
[ 2. 2. 2. ... 2. 2. 2.]
...
[597. 597. 597. ... 597. 597. 597.]
[598. 598. 598. ... 598. 598. 598.]
[599. 599. 599. ... 599. 599. 599.]]
