def blur():
image = cv2.imread(IMG_PATH + "/yumi-cells.jpg")
kernel_sizes = [(3, 3), (5, 5), (7, 7), (7, 1), (1, 7)]
filter_imgs = {"original": image}
blur_imgs = {"original": image}
for ksize in kernel_sizes:
title = f"ksize: {ksize}"
kernel = np.ones(ksize)
kernel /= kernel.size
filter_imgs[title] = cv2.filter2D(image, -1, kernel)
blur_imgs[title] = cv2.blur(image, ksize)
resimg = si.show_imgs(filter_imgs, "cv2.filter2D", 3)
resimg = si.show_imgs(blur_imgs, "cv2.blur", 3)
def watershed():
srcimg = cv2.imread(IMG_PATH + "/wildgoose.jpg", cv2.IMREAD_COLOR)
images = {"original": srcimg}
# find apprent background(sky) and wild goose region
gray = cv2.cvtColor(srcimg, cv2.COLOR_BGR2GRAY)
gray = cv2.GaussianBlur(gray, (3, 3), 0)
ret, binary = cv2.threshold(gray, 150, 255, cv2.THRESH_BINARY)
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (3, 3))
images["threshold"] = binary
app_sky = cv2.erode(binary, kernel, iterations=3)
images["apparent sky"] = app_sky
app_goose = cv2.dilate(binary, kernel, iterations=1)
images["apparent wildgoose"] = app_goose
images["undetermined"] = cv2.subtract(app_goose, app_sky)
# create markers and watershed
markers = np.zeros(gray.shape, dtype=np.int32)
markers[app_sky > 0] = SKY_LABEL
markers[app_goose == 0] = GOOSE_LABEL
markers = cv2.watershed(srcimg, markers)
# show contours of wild gooses
labelimg = np.zeros_like(srcimg)
labelimg[markers == -1, :] = (0, 0, 255) # draw contours
labelimg[markers == SKY_LABEL, :] = (200, 150, 100)
labelimg[markers == GOOSE_LABEL, :] = srcimg[markers == GOOSE_LABEL, :]
images["label"] = labelimg
result_img = si.show_imgs(images, "watershed", 3)
cv2.imwrite(IMG_PATH + "/result/watershed_wildgoose.jpg", result_img)
def count_puzzle(img):
ih, iw, ch = img.shape
mask = np.zeros((ih+2, iw+2), dtype=np.uint8)
images = {"original": img.copy()}
cv2.imshow("image", img)
cv2.waitKey()
count = 0
for y in range(img.shape[0]):
for x in range(img.shape[1]):
if mask[y+1, x+1] == 0:
# flood fill로 채울 랜덤 색상 생성
color = np.random.randint(50, 256, 3).tolist()
color[0] = 255
ret, img, mask, rect = cv2.floodFill(img, mask, (x,y), color, (10,10,10), (10,10,10), flags=8)
mask_show = mask*255
print(f"area={ret}, rect={rect}, mask value={mask[y+1, x+1]}")
if ret > 500: # 영역이 넓은 것만 퍼즐 조각으로 인정
cv2.imshow("image", img)
cv2.imshow("mask", mask_show)
cv2.waitKey(200)
count += 1
print("total puzzle count:", count)
cv2.destroyAllWindows()
images["filled"] = img
result_img = si.show_imgs(images, "floodfill", 2)
cv2.imwrite(IMG_PATH + "/result/floodfill_puzzle.jpg", result_img)
def canny():
image = cv2.imread(IMG_PATH + "/arya.jpg", cv2.IMREAD_GRAYSCALE)
canny_imgs = {"original": image}
canny_imgs["Laplacian"] = cv2.Laplacian(image, -1, scale=2)
canny_imgs["Canny (100, 200)"] = cv2.Canny(image, 100, 200)
canny_imgs["Canny (200, 255)"] = cv2.Canny(image, 150, 255)
result_img = si.show_imgs(canny_imgs, "Canny", 2)
def fill_lake():
srcimg = cv2.imread(IMG_PATH + "/sinjeong-lake.jpg", cv2.IMREAD_COLOR)
images = {"original": srcimg}
seed = (int(srcimg.shape[1] / 2), int(srcimg.shape[0] / 2))
# 아무 처리하지 않고 바로 호수에 flood fill 적용
fill_direct = srcimg.copy()
mask = np.zeros((srcimg.shape[0] + 2, srcimg.shape[1] + 2), dtype=np.uint8)
retval, fill_direct, mask, rect = cv2.floodFill(fill_direct,
mask,
seed, (0, 255, 255),
(2, 2, 2), (2, 2, 2),
flags=8)
print(f"pixel area of lake WITHOUT preprocess={retval}, rect={rect}")
fill_direct = cv2.circle(fill_direct, seed, 1, (0, 0, 255), 2)
images["direct_floodfill"] = fill_direct
# flood fill이 잘 퍼지도록 블러링 후 적용
fill_blur = srcimg.copy()
fill_blur = cv2.GaussianBlur(fill_blur, (3, 3), 0)
fill_blur = cv2.medianBlur(fill_blur, 3)
mask = np.zeros((srcimg.shape[0] + 2, srcimg.shape[1] + 2), dtype=np.uint8)
retval, fill_blur, mask, rect = cv2.floodFill(fill_blur,
mask,
seed, (0, 255, 255),
(2, 2, 2), (2, 2, 2),
flags=8 | (255 << 8))
print(f"pixel area of lake WITH preprocess= {retval}, rect={rect}")
fill_blur = cv2.circle(fill_blur, seed, 1, (0, 0, 255), 2)
images["blur_n_floodfill"] = fill_blur
# 결과 출력
images["final mask"] = cv2.cvtColor(mask[1:-1, 1:-1], cv2.COLOR_GRAY2BGR)
result_img = si.show_imgs(images, "fill the lake", 2)
cv2.imwrite(IMG_PATH + "/result/fill_lake.jpg", result_img)
def gaussian():
image = cv2.imread(IMG_PATH + "/yumi-cells.jpg")
kernel_size = (5, 5)
blur_imgs = {}
blur_imgs["original"] = image
blur_imgs["blur"] = cv2.blur(image, kernel_size)
blur_imgs["GaussianBlur"] = cv2.GaussianBlur(image, kernel_size, 0)
result_img = si.show_imgs(blur_imgs, "GaussianBlur", 3, 1000)
def laplacian():
image = cv2.imread(IMG_PATH + "/yumi-cells.jpg", cv2.IMREAD_GRAYSCALE)
lapla_imgs = {"original": image}
sobel_dx = cv2.Sobel(image, ddepth=-1, dx=1, dy=0, ksize=3)
sobel_dy = cv2.Sobel(image, ddepth=-1, dx=0, dy=1, ksize=3)
lapla_imgs["Sobel dx+dy"] = cv2.add(sobel_dx, sobel_dy)
lapla_imgs["Laplacian"] = cv2.Laplacian(image, -1)
result_img = si.show_imgs(lapla_imgs, "Laplacian", 3)
def erode_and_dilate():
srcimg = cv2.imread(IMG_PATH + "/marvel.jpg", cv2.IMREAD_GRAYSCALE)
srcimg = salt_and_pepper_noise(srcimg)
# erode and dilate, and then show them
images = {"original": srcimg}
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (3, 3))
images["erode"] = cv2.erode(srcimg, kernel)
images["dilate"] = cv2.dilate(srcimg, kernel)
result_img = si.show_imgs(images, "Erode and Dilate", 1)
def median():
image = cv2.imread(IMG_PATH + "/saltnpepper.jpg")
kernel_size = (5, 5)
median_imgs = dict()
median_imgs["original"] = image
median_imgs[f"gaussian {kernel_size}"] = cv2.GaussianBlur(
image, kernel_size, 0)
median_imgs[f"median {kernel_size[0]}"] = cv2.medianBlur(
image, kernel_size[0])
result_img = si.show_imgs(median_imgs, "Median Filter", 3)
# cv2.imwrite(IMG_PATH + "/result/snp-median.jpg", result_img)
image = cv2.imread(IMG_PATH + "/ann.jpg")
median_imgs = dict()
median_imgs["original"] = image
median_imgs["median (3)"] = cv2.medianBlur(image, 3)
median_imgs["median (5)"] = cv2.medianBlur(image, 5)
result_img = si.show_imgs(median_imgs, "Median Filter", 3)
def sobel():
image = cv2.imread(IMG_PATH + "/yumi-cells.jpg", cv2.IMREAD_GRAYSCALE)
sobel_imgs = {"original": image}
sobel_imgs["Sobel dx"] = cv2.Sobel(image, ddepth=-1, dx=1, dy=0, ksize=3)
sobel_imgs["Sobel dy"] = cv2.Sobel(image, ddepth=-1, dx=0, dy=1, ksize=3)
sobel_imgs["Sobel dx+dy"] = cv2.add(sobel_imgs["Sobel dx"],
sobel_imgs["Sobel dy"])
sobel_imgs["Scharr dx"] = cv2.Scharr(image, ddepth=-1, dx=1, dy=0)
sobel_imgs["Scharr dy"] = cv2.Scharr(image, ddepth=-1, dx=0, dy=1)
result_img = si.show_imgs(sobel_imgs, "Sobel & Scharr", 3)
def count_balls():
srcimg = cv2.imread(IMG_PATH + "/ballpool.jpg", cv2.IMREAD_COLOR)
images, mask = prepare_mask(srcimg)
result_img = si.show_imgs(images, "floodfill", 3)
label = LABEL_BEGIN
ih, iw, ch = srcimg.shape
hueimg = images["hue"].copy()
for v in range(0, ih, 5):
for u in range(0, iw, 5):
if mask[v+1, u+1] == 0:
hueimg, mask, area = fill_image(hueimg, mask, (u, v), label)
cv2.imshow("floodfill mask", mask)
if area > AREA_THR:
label += 1
cv2.waitKey(100)
cv2.waitKey()
labeled = colorize_regions(mask, label)
images["labeled balls"] = labeled
result_img = si.show_imgs(images, "floodfill", 3)
cv2.imwrite(IMG_PATH + "/result/floodfill_ball.jpg", result_img)
def bilateral():
image = cv2.imread(IMG_PATH + "/road.jpg")
kernel_size = (5, 5)
blur_imgs = {}
blur_imgs["original"] = image
blur_imgs["gaussian"] = cv2.GaussianBlur(image, kernel_size, 0)
blur_imgs["bilateral (5,50,50)"] = cv2.bilateralFilter(image, 5, 50, 50)
blur_imgs["bilateral (5,150,150)"] = cv2.bilateralFilter(
image, 5, 150, 150)
result_img = si.show_imgs(blur_imgs, "Bilateral Filter", 2)
cv2.imwrite(IMG_PATH + "/result/road-bilateral.jpg", result_img)
def morphologies():
srcimg = cv2.imread(IMG_PATH + "/marvel.jpg", cv2.IMREAD_GRAYSCALE)
srcimg = salt_and_pepper_noise(srcimg)
# erode and dilate, and then show them
images = {"original": srcimg}
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (3, 3))
images["opening"] = cv2.morphologyEx(srcimg, cv2.MORPH_OPEN, kernel)
images["closing"] = cv2.morphologyEx(srcimg, cv2.MORPH_CLOSE, kernel)
images["gradient"] = cv2.morphologyEx(srcimg, cv2.MORPH_GRADIENT, kernel)
images["tophat"] = cv2.morphologyEx(srcimg, cv2.MORPH_TOPHAT, kernel)
images["blackhat"] = cv2.morphologyEx(srcimg, cv2.MORPH_BLACKHAT, kernel)
result_img = si.show_imgs(images, "Morphology Ops", 2)