def calc_slopes(img, chunks, lines):
chunk_size = height(img) // chunks
slopes = [[0 for _ in range(width(img))] for c in range(chunks)]
for i in range(width(img)):
s = [0 for _ in range(chunks)]
num = [0 for _ in range(chunks)]
for line in lines:
x1,y1,x2,y2 = line[0]
if (x1 <= i and x2 >= i) or (x1 <= i and x2 >= i):
for chunk in range(chunks):
c1 = y1 // chunk_size
c2 = y2 // chunk_size
if (c1 >= chunk and c2 <= chunk) or (c2 >= chunk and c1 <= chunk):
s[chunk] += float(y2 - y1) / float(x2 - x1)
num[chunk] += 1
for j in range(chunks):
if num[j] > 0:
slopes[j][i] = float(s[j]) / float(num[j])
img_blank = img.copy()
k = 0
for s in slopes:
row = int(np.round((height(img_blank) // len(slopes)) * (k + .5)))
k += 1
line = calc_staff_line(s, row, img)
for pt in line:
img_blank[pt[0]][pt[1]] = 255
return slopes, img_blank
def calc_staff_line_chunks(slopes, row, img, chunk_size, interval = 15):
white_seen = 0
col = 0
r = row
line = []
for col in range(0, width(img), interval):
if (int(np.round(r)) < height(img) and col < width(img)):
for i in range(interval):
line.append((int(np.round(r)), col + i))
if(img[int(np.round(r))][col] == 255):
white_seen += 1
if (int(r // chunk_size) < len(slopes) and int(r // chunk_size) >= 0):
r += slopes[int(r // chunk_size)][col] * float(interval)
col += 1
return line, white_seen*interval
def find_staff_candidates(img, slopes, T_staff_cand, T_length, num_chunks, staff_height, staff_space, min_y, max_y):
staff_candidates = []
chunk_size = height(img) // num_chunks
rows = [i for i in range(min_y - 50, max_y + 50)]
for row in rows:
pts, total_white = calc_staff_line_chunks(slopes, row, img, chunk_size)
if total_white > 0:
if (float(total_white) / width(img) > T_staff_cand):
if (len(staff_candidates) == 0
or (len(staff_candidates) > 0 and row - staff_candidates[-1][0] > T_length)):
staff_candidates.append((row, pts, total_white))
if (len(staff_candidates) > 0 and row - staff_candidates[-1][0] <= T_length and total_white > staff_candidates[-1][2]):
staff_candidates[len(staff_candidates) - 1] = (row, pts, total_white)
to_remove = []
T_staff = (5 * staff_height) + (7 * staff_space)
curr = []
for cand in staff_candidates:
curr.append(cand)
if (curr[-1][0] - curr[0][0] > T_staff):
curr = curr[:len(curr) - 1]
while (len(curr) > 5):
tr = [x for x in curr if x[2] == min([c[2] for c in curr])][0]
to_remove.append(tr)
curr = [x for x in curr if x != tr]
curr = []
for r in to_remove:
staff_candidates.remove(r)
print([str(x[0]) + " " + str(x[2]) for x in staff_candidates])
return staff_candidates
def calc_staff_line(slopes, row, img):
col = 0
r = row
line = []
for slope in slopes:
if (int(np.round(r)) < height(img) and col < width(img)):
line.append((int(np.round(r)), col))
r += slope
col += 1
return line
def remove_text(img):
new_img = img.copy()
kernel = cv.getStructuringElement(cv.MORPH_CROSS, (3, 3))
dilated = cv.dilate(new_img, kernel, iterations=5)
_, contours, hierarchy = cv.findContours(dilated, cv.RETR_EXTERNAL,
cv.CHAIN_APPROX_NONE)
for contour in contours:
[x, y, w, h] = cv.boundingRect(contour)
if (w < width(img) // 2):
cv.rectangle(img, (x, y), (x + w, y + h), (0, 0, 0), cv.FILLED)
return img
def estimate_staff_stats(img):
white_runs = []
black_runs = []
for i in range(0, width(img), 20):
wr, br = calc_runs(img[:, i], 255, 0)
white_runs.extend(wr)
black_runs.extend(br)
staff_space = int(np.median(black_runs))
staff_height = int(np.median(white_runs)) #image is bit flipped
T_length = int(min(2.0 * staff_height, staff_height + staff_space))
return staff_height, staff_space, T_length
def create_initial_staff_image(img, T_length):
img_copy = img.copy()
for col in range(width(img_copy)):
i = 0
while(i < height(img_copy)):
while(i < height(img_copy) and img_copy[i][col] == 0):
i+=1
mark = i
while(i < height(img_copy) and img_copy[i][col] == 255):
i+=1
if (i - mark > T_length):
for j in range(mark, i):
img_copy[j][col] = 0
return img_copy