python Sobelフィルタ + アンシャープマスク（先鋭化）

🔳用途

・画像の、エッジ部分を強調して出す。
・線形のフィルタリングなので、Sobelフィルタ単体よりも、よりエッジの値が上がる。

🔳結果

・フィルタなし画像

・フィルタあり画像

🔳ソースコード

# ================================
# 1セル完結テンプレ（Colab）
# Driveマウント -> 画像選択 -> 表示 -> Sobel -> Unsharp -> Sobel(先鋭後)
# OpenCVなし / NumPy + PIL + matplotlib + ipywidgets
# ================================

#--- install
!pip -q install ipywidgets

import os, glob 
import numpy as np
from PIL import Image
import matplotlib.pyplot as plt
import ipywidgets as widgets
from IPython.display import display, clear_output

from google.colab import drive
drive.mount('/content/drive')

# ========= ここを必要なら変更 =========
# Drive内のどこから探すか（重いならフォルダを絞る）
# ROOT = "/content/drive/MyDrive"
ROOT = "/content/drive/MyDrive/画像フィルタ_25-26_001"
# =====================================

# ---------- 画像処理 基本関数 ----------
def to_gray_np(img_np):
  """img_np: (H,W,3) or (H,W) 0..255 -> float32 gray"""
  if img_np.ndim == 2:
    return img_np.astype(np.float32)
  r, g, b = img_np[..., 0], img_np[..., 1], img_np[..., 2]
  return (0.299*r + 0.587*g + 0.114*b).astype(np.float32)

def conv2d_gray(gray, kernel):
  gray = gray.astype(np.float32)
  k = np.array(kernel, dtype=np.float32)
  kh, kw = k.shape
  pad_h, pad_w = kh//2, kw//2
  padded = np.pad(gray, ((pad_h, pad_h), (pad_w, pad_w)), mode="edge")
  out = np.zeros_like(gray, dtype=np.float32)

  for y in range(out.shape[0]):
    for x in range(out.shape[1]):
      region = padded[y:y+kh, x:x+kw]
      out[y, x] = np.sum(region * k)
  return out

def sobel(gray):
  """gray: (H,W) float32"""
  Kx = np.array([[-1, 0, 1],
                 [-2, 0, 2],
                 [-1, 0, 1]], dtype=np.float32)
  Ky = np.array([[1, 2, 1],
                 [0, 0, 0],
                 [-1, -2, -1]], dtype=np.float32)
  gx = conv2d_gray(gray, Kx)
  gy = conv2d_gray(gray, Ky)
  mag = np.sqrt(gx*gx + gy*gy)
  mag = mag / (mag.max() + 1e-8) * 255.0
  return gx, gy,mag.astype(np.uint8)

def gaussian_kernel_1d(radius, sigma):
  x = np.arange(-radius, radius+1, dtype=np.float32)
  k = np.exp(-(x*x)/(2*sigma*sigma))
  k /= (k.sum() + 1e-8)
  return k

def conv1d_h(gray, k):
  r = len(k) // 2
  padded = np.pad(gray, ((0,0),(r,r)), mode="edge")
  out = np.zeros_like(gray, dtype=np.float32)
  for y in range(gray.shape[0]):
    for x in range(gray.shape[1]):
      out[y,x] = np.sum(padded[y, x:x+2*r+1] * k)
  return out

def conv1d_v(gray, k):
    r = len(k)//2
    padded = np.pad(gray, ((r,r),(0,0)), mode="edge")
    out = np.zeros_like(gray, dtype=np.float32)
    for y in range(gray.shape[0]):
        for x in range(gray.shape[1]):
            out[y,x] = np.sum(padded[y:y+2*r+1, x] * k)
    return out

def gaussian_blur(gray, radius=2, sigma=1.2):
    k = gaussian_kernel_1d(radius, sigma)
    tmp = conv1d_h(gray.astype(np.float32), k)
    out = conv1d_v(tmp, k)
    return out

def unsharp_mask(gray, radius=2, sigma=1.2, amount=1.2, threshold=2.0):
    g = gray.astype(np.float32)
    blur = gaussian_blur(g, radius=radius, sigma=sigma)
    detail = g - blur
    if threshold > 0:
        mask = np.abs(detail) >= threshold
        detail = detail * mask
    sharp = g + amount * detail
    sharp = np.clip(sharp, 0, 255)
    return sharp.astype(np.uint8), blur.astype(np.uint8)

def show_results(img_rgb, gray, edge, blur, sharp, edge_sharp, title_path=""):
    h, w = img_rgb.shape[:2]
    disp_w = min(900, w)
    disp_h = int(disp_w * h / w)
    img_disp = Image.fromarray(img_rgb).resize((disp_w, disp_h))

    plt.figure(figsize=(18,5))
    plt.subplot(1,5,1); plt.title("Original"); plt.imshow(img_disp); plt.axis("off")
    plt.subplot(1,5,2); plt.title("Gray"); plt.imshow(gray, cmap="gray"); plt.axis("off")
    plt.subplot(1,5,3); plt.title("Sobel"); plt.imshow(edge, cmap="gray"); plt.axis("off")
    plt.subplot(1,5,4); plt.title("Unsharp"); plt.imshow(sharp, cmap="gray"); plt.axis("off")
    plt.subplot(1,5,5); plt.title("Sobel on Unsharp"); plt.imshow(edge_sharp, cmap="gray"); plt.axis("off")
    plt.suptitle(title_path, y=1.02, fontsize=11)
    plt.show()

# ---------- Drive内の画像を収集 ----------
exts = ("*.png","*.jpg","*.jpeg","*.bmp","*.webp")
paths = []
for e in exts:
  paths += glob.glob(os.path.join(ROOT, "**", e), recursive=True)
paths = sorted(paths)

if len(paths) == 0:
  raise FileNotFoundError(f"画像が見つからないよ。ROOT={ROOT} を画像フォルダに変更してね。")

MAX_LIST = 2000
paths_view = paths[:MAX_LIST] 

# ---------- UI ----------
dd = widgets.Dropdown(
    options=paths_view,
    description="Image:",
    layout=widgets.Layout(width="95%")
)

radius = widgets.IntSlider(value=2, min=1, max=8, step=1, description="radius")
sigma  = widgets.FloatSlider(value=1.2, min=0.5, max=5.0, step=0.1, description="sigma")
amount = widgets.FloatSlider(value=1.2, min=0.0, max=3.0, step=0.1, description="amount")
thresh = widgets.FloatSlider(value=2.0, min=0.0, max=20.0, step=0.5, description="threshold")

btn_run = widgets.Button(description="Run", button_style="success")
out = widgets.Output()

def run_process(_):
    with out:
        clear_output(wait=True)
        path = dd.value
        img = np.array(Image.open(path).convert("RGB"))
        gray = to_gray_np(img)

        _, _, edge = sobel(gray)
        sharp, blur = unsharp_mask(gray, radius=radius.value, sigma=sigma.value, amount=amount.value, threshold=thresh.value)
        _, _, edge_sharp = sobel(sharp.astype(np.float32))

        show_results(img, gray, edge, blur, sharp, edge_sharp, title_path=path)
        print(f"Loaded: {path}")
        print(f"Params: radius={radius.value}, sigma={sigma.value}, amount={amount.value}, threshold={thresh.value}")
        if len(paths) > MAX_LIST:
            print(f"※画像が多いので表示候補は先頭 {MAX_LIST} 件だけ。ROOTを絞ると探しやすいよ。")

btn_run.on_click(run_process)

display(dd, widgets.HBox([btn_run]), widgets.VBox([radius, sigma, amount, thresh]), out)

# 初回自動実行
run_process(None)