def convolve(img, filters, show=False):
accum = np.zeros_like(img)
for filter in filters:
img_filt = cv2.filter2D(img, cv2.CV_8UC3, filter)
# img_filt += img_filt
np.maximum(accum, img_filt, accum)
if show:
titles = ['Original', 'Filtered']
images = [img, img_filt]
title = 'Apply Gabor filter'
tl.plotImages(titles, images, title, 1, 2)
return accum
def detectDefects(img_filtered, show=False):
kernel_close = np.ones((5, 5), np.uint8)
kernel_dilate = np.ones((5, 5), np.uint8)
defects_detected_nms = None
copy = img_filtered.copy()
# Threshold image
thresh = cv2.threshold(copy, 65, 220, cv2.THRESH_BINARY_INV)[1]
# Clean image
thresh = cv2.morphologyEx(thresh,
cv2.MORPH_CLOSE,
kernel_close,
iterations=3)
thresh = cv2.dilate(thresh, kernel_dilate, iterations=2)
# Find contours
contours = cv2.findContours(thresh, cv2.RETR_EXTERNAL,
cv2.CHAIN_APPROX_SIMPLE)[1]
contours = [
cnt for cnt in contours
if cv2.boundingRect(cnt)[2] > 10 and cv2.boundingRect(cnt)[3] > 10
]
# filtered_contours = []
# for cnt in contours:
# if cnt[2] > 10 and cnt[3] > 10:
# filtered_contours.append(cnt)
if len(contours) != 0:
# Get all bboxes
defects_detected = [cv2.boundingRect(contour) for contour in contours]
# Apply NMS to all bboxes
defects_detected_nms = non_max_suppression_fast(
np.array(defects_detected), 0.1)
if show:
copy_th = cv2.cvtColor(thresh.copy(), cv2.COLOR_GRAY2BGR)
# copy_th = cv2.drawContours(copy_th, contours, -1, (0, 255, 0), 2)
for (x, y, w, h) in defects_detected_nms:
cv2.rectangle(copy_th, (x, y), (x + w, y + h), (255, 0, 0), 2)
titles = ['Filtered', 'Thresh (bbox)']
images = [img_filtered, copy_th]
title = 'Detect defects'
tl.plotImages(titles, images, title, 1, 2)
return defects_detected_nms
def removeBG(img, margin, show=False):
img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
img = cv2.GaussianBlur(img, (5, 5), 0)
thresh = cv2.threshold(img, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)[1]
contours = cv2.findContours(thresh, cv2.RETR_EXTERNAL,
cv2.CHAIN_APPROX_SIMPLE)[1]
if len(contours) != 0:
# find the biggest area
cnt_wood = max(contours, key=cv2.contourArea)
x, y, w, h = cv2.boundingRect(cnt_wood)
fg = img[y + margin:y + h - margin, x:x + w]
if show:
copy = cv2.cvtColor(img.copy(), cv2.COLOR_GRAY2BGR)
cv2.rectangle(copy, (x, y), (x + w, y + h), (0, 255, 255), 2)
titles = ['Contour', 'ROI']
images = [copy, fg]
title = 'Remove background'
tl.plotImages(titles, images, title, 1, 2)
return fg, (x, y, w, h)
def build_filters(size, theta, sigma, show=False):
filters = []
ksize = size
for t in theta:
for s in sigma:
#cv2.getGaborKernel(ksize, sigma, theta, lambda, gamma, psi, ktype)
kern = np.real(
cv2.getGaborKernel((ksize, ksize),
s,
t,
10.0,
0.5,
0,
ktype=cv2.CV_32F))
kern /= 1.5 * kern.sum()
filters.append(kern)
if show:
titles = []
images = np.real(filters)
title = 'Gabor kernels'
tl.plotImages(titles, images, title, 6, 8)
return filters
margin)
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
# Draw GT
if gt is not None:
for x, y, w, h in gt:
cv2.rectangle(img, (x, y), (x + w, y + h), (0, 255, 0), 2)
# Draw detections
if defects_detected is not None:
for x, y, w, h in defects_detected:
cv2.rectangle(img, (x, y), (x + w, y + h), (255, 0, 0), 2)
if show:
titles = []
images = [img]
title = "Detection result for '{}'".format(file)
tl.plotImages(titles, images, title, 1, 1)
results.append(imp.evaluate(gt, defects_detected)) # debug
prev_file = substr
# break
results = np.array(results)
iou_avg = np.mean([np.mean(iou) for iou in results[:, 0]])
TR = np.sum(results[:, 1]).astype(int)
FR = np.sum(results[:, 2]).astype(int)
FA = np.sum(results[:, 3]).astype(int)
TA = np.sum(results[:, 4]).astype(int)
print(text.format(iou_avg, TR, FR, FA, TA))
classes = ['rejected', 'accepted']
cnf_matrix = np.zeros((2, 2), dtype=int)