Morfologiya: shaklni yupqalashtirish, yo'g'onlashtirish, tozalash

img = cv2.imread(str(RASM / "shovqinli.png"), cv2.IMREAD_GRAYSCALE)
_, binar = cv2.threshold(img, 200, 255, cv2.THRESH_BINARY_INV)          # shakllar oq bo'lsin
yadro = np.ones((5, 5), np.uint8)
eroziya = cv2.erode(binar, yadro, iterations=1)
kengayish = cv2.dilate(binar, yadro, iterations=1)
print(f"oq piksellar — asl: {int((binar > 0).sum())}, erosion: {int((eroziya > 0).sum())}, "
      f"dilation: {int((kengayish > 0).sum())}")
ko_rsat(binar, eroziya, kengayish, nom="erode_dilate.png",
        sarlavhalar=["binar (shovqinli)", "erosion 5x5", "dilation 5x5"],
        izoh="Erosion shovqin nuqtalarini yo'q qildi, lekin shakllarni ham kichraytirdi. Dilation teskarisi.")

ochish = cv2.morphologyEx(binar, cv2.MORPH_OPEN, yadro)
yopish = cv2.morphologyEx(binar, cv2.MORPH_CLOSE, yadro)
ikkisi = cv2.morphologyEx(ochish, cv2.MORPH_CLOSE, yadro)
toza = cv2.imread(str(RASM / "shakllar.png"), cv2.IMREAD_GRAYSCALE)
_, haqiqat = cv2.threshold(toza, 200, 255, cv2.THRESH_BINARY_INV)

def moslik(a):
    """Haqiqat bilan piksel darajasida moslik: IoU."""
    kesishma = int(((a > 0) & (haqiqat > 0)).sum())
    birlashma = int(((a > 0) | (haqiqat > 0)).sum())
    return round(100 * kesishma / birlashma, 1)

show(pd.DataFrame([{"usul": n, "oq piksellar": int((v > 0).sum()), "haqiqatga moslik (IoU) %": moslik(v)}
                   for n, v in [("binar (xom)", binar), ("opening", ochish), ("closing", yopish),
                                ("opening + closing", ikkisi)]]))
ko_rsat(ochish, yopish, ikkisi, nom="open_close.png",
        sarlavhalar=["opening — nuqtalar ketdi", "closing — teshiklar to'ldi", "ikkisi ketma-ket"],
        izoh="Tuz-qalampir shovqinida opening kerak; teshikli shaklda closing. Ko'pincha ikkalasi ketma-ket.")

shakllar_yadro = {"RECT": cv2.MORPH_RECT, "ELLIPSE": cv2.MORPH_ELLIPSE, "CROSS": cv2.MORPH_CROSS}
qatorlar = []
for nom, tur in shakllar_yadro.items():
    k = cv2.getStructuringElement(tur, (9, 9))
    v = cv2.morphologyEx(binar, cv2.MORPH_OPEN, k)
    qatorlar.append({"yadro": nom, "yadro ichidagi birlar": int(k.sum()),
                     "oq piksellar": int((v > 0).sum()), "IoU %": moslik(v)})
show(pd.DataFrame(qatorlar))
print("ELLIPSE yadro (9x9):\n", cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (9, 9)))

hujjat = cv2.imread(str(RASM / "hujjat.jpg"), cv2.IMREAD_GRAYSCALE)
P = cv2.getPerspectiveTransform(np.float32([[150, 90], [720, 170], [650, 940], [90, 830]]),
                                np.float32([[0, 0], [640, 0], [640, 900], [0, 900]]))
togri = cv2.warpPerspective(hujjat, P, (640, 900))
matn = cv2.adaptiveThreshold(togri, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY_INV, 31, 15)
qatorlar = []
variantlar = {}
for kenglik in (1, 5, 15, 40, 120):
    if kenglik == 1:
        chiqdi = matn
    else:
        chiqdi = cv2.dilate(matn, cv2.getStructuringElement(cv2.MORPH_RECT, (kenglik, 3)))
    variantlar[kenglik] = chiqdi
    qatorlar.append({"yadro kengligi": kenglik, "alohida bo'laklar": cv2.connectedComponents(chiqdi)[0] - 1})
show(pd.DataFrame(qatorlar))
ko_rsat(variantlar[1][:420], variantlar[15][:420], variantlar[120][:420], nom="matn_morfologiya.png",
        sarlavhalar=["yadrosiz — harflar", "15 px — so'zlar", "120 px — butun qator"],
        izoh="Gorizontal yadroni kengaytirsangiz harf -> so'z -> ustun -> qator -> paragraf bo'lib birlashadi.")

gradient = cv2.morphologyEx(haqiqat, cv2.MORPH_GRADIENT, yadro)
tophat = cv2.morphologyEx(togri, cv2.MORPH_TOPHAT, cv2.getStructuringElement(cv2.MORPH_RECT, (31, 31)))
blackhat = cv2.morphologyEx(togri, cv2.MORPH_BLACKHAT, cv2.getStructuringElement(cv2.MORPH_RECT, (31, 31)))
ko_rsat(gradient, tophat, blackhat, nom="gradient_tophat.png",
        sarlavhalar=["GRADIENT (dilation − erosion)", "TOPHAT", "BLACKHAT"],
        izoh="Gradient — shaklning konturi. Blackhat qorong'i tafsilotni (matnni) yorug' fondan ajratadi.")