| import io |
|
|
| import numpy as np |
| from PIL import Image, ImageFilter |
|
|
|
|
| def features(image: bytes) -> dict: |
| with Image.open(io.BytesIO(image)) as im: |
| gray = im.convert("L").resize((256, 256)) |
| a = np.asarray(gray, dtype=float) |
| ink = a < 180 |
| contrast = float(a.std()) |
| edge = float(np.asarray(gray.filter(ImageFilter.FIND_EDGES), dtype=float).var()) |
| horizontal = int((ink.mean(axis=1) > 0.7).sum()) |
| vertical = int((ink.mean(axis=0) > 0.7).sum()) |
| |
| density = float(ink.mean()) |
| ink_contrast = contrast / max(0.001, density**0.5) |
| return dict( |
| blank=density < 0.0001, |
| contrast=contrast, |
| ink_contrast=ink_contrast, |
| edge_variance=edge, |
| line_density=density, |
| table_rulings=horizontal + vertical, |
| possible_columns=bool( |
| ink[:, :100].mean() > 0.03 |
| and ink[:, 156:].mean() > 0.03 |
| and ink[:, 120:136].mean() < 0.01 |
| ), |
| complex=horizontal + vertical > 8 or (ink_contrast < 145 and density > 0.001), |
| handwriting_proxy="unknown", |
| orientation="landscape" if im.width > im.height else "portrait", |
| ) |
|
|
