computer-vision

Files (2)

• SKILL.md
• metadata.json

SKILL.md (first lines)

# Computer Vision

## When to Use
Use this skill for image loading/manipulation, object detection, OCR text extraction, image classification, manga panel/bubble processing, or building batch image pipelines. Covers PIL/Pillow, OpenCV, Tesseract, EasyOCR, YOLO, and Claude Vision API.

---

## Core Rules
- PIL/Pillow for image I/O and basic transforms; OpenCV for pixel-level operations and contour detection.
- For OCR quality: Claude Vision API > EasyOCR > Tesseract (especially for manga/stylized fonts).
- Always convert PIL images to numpy arrays for OpenCV (`np.array(img)`), and back with `Image.fromarray(arr)`.
- BGR vs RGB: OpenCV loads as BGR; PIL/most APIs expect RGB. Always convert: `cv2.cvtColor(img, cv2.COLOR_BGR2RGB)`.
- Batch pipelines: use `pathlib.Path` + generator patterns for memory efficiency on large sets.

---

## PIL / Pillow Basics

```python
from PIL import Image, ImageFilter, ImageEnhance, ImageDraw
import numpy as np

# Load, inspect, save
img = Image.open("photo.jpg")
print(img.size, img.mode)   # (width, height), 'RGB' / 'RGBA' / 'L'
img.save("output.png")

# Resize
img_resized = img.resize((800, 600))
img_thumb = img.copy(); img_thumb.thumbnail((256, 256))  # preserves aspect

# Convert modes
gray = img.convert("L")       # grayscale
rgba = img.convert("RGBA")    # add alpha channel

# Crop (left, upper, right, lower)
cropped = img.crop((100, 100, 500, 400))

# Rotate / flip
rotated = img.rotate(90, expand=True)
flipped = img.transpose(Image.FLIP_LEFT_RIGHT)

# Filters
blurred = img.filter(ImageFilter.GaussianBlur(radius=3))
sharp = img.filter(ImageFilter.SHARPEN)

# Enhance
enhancer = ImageEnhance.Contrast(img)
img_high_contrast = enhancer.enhance(2.0)

# Draw on image
draw = ImageDraw.Draw(img)
draw.rectangle([10, 10, 200, 200], outline="red", width=3)
draw.text((20, 20), "Label", fill="white")

# PIL <-> numpy
arr = np.array(img)          # PIL to numpy (RGB)
img_back = Image.fromarray(arr)  # numpy to PIL
```

---

## OpenCV Fundamentals

```python
import cv2
import numpy as np

# Read / write
img = cv2.imread("photo.jpg")          # BGR format
img_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)  # fix to RGB
cv2.imwrite("output.jpg", img)

# Resize
img_resized = cv2.resize(img, (800, 600))
img_half = cv2.resize(img, None, fx=0.5, fy=0.5)

# Grayscale + threshold
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
_, binary = cv2.threshold(gray, 128, 255, cv2.THRESH_BINARY)
adaptive = cv2.adaptiveThreshold(gray, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
                                  cv2.THRESH_BINARY, 11, 2)

# Blur
blurred = cv2.GaussianBlur(img, (5, 5), 0)
median = cv2.medianBlur(img, 5)

# Edge detection
edges = cv2.Canny(gray, 50, 150)

# Contour detection (e.g. panel borders)
contours, hierarchy = cv2.findContours(binary, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
for cnt in contours:
    area = cv2.contourArea(cnt)
    if area > 5000:
        x, y, w, h = cv2.boundingRect(cnt)
        cv2.rectangle(img, (x, y), (x+w, y+h), (0, 255, 0), 2)

# Morphological ops (useful for manga cleanup)
kernel = np.ones((3, 3), np.uint8)
dilated = cv2.dilate(binary, kernel, iterations=1)
eroded = cv2.erode(binary, kernel, iterations=1)
cleaned = cv2.morphologyEx(binary, cv2.MORPH_CLOSE, kernel)

# Display (non-blocking for scripts)
cv2.imshow("Result", img)
cv2.waitKey(0)
cv2.destroyAllWindows()
```

---

## OCR with Tesseract (pytesseract)

```bash
# Install
pip install pytesseract pillow
brew install tesseract               # macOS
# sudo apt install tesseract-ocr    # Linux

From UltronCore/claude-skill-vault, MIT.

More from UltronCore/claude-skill-vault