File size: 14,356 Bytes
b4123b8 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 |
"""
Texture feature extraction for the Sorghum Pipeline.
This module handles extraction of texture features including:
- Local Binary Patterns (LBP)
- Histogram of Oriented Gradients (HOG)
- Lacunarity features
- Edge Histogram Descriptor (EHD)
"""
import numpy as np
import cv2
import torch
import torch.nn.functional as F
from skimage.feature import local_binary_pattern, hog
from skimage import exposure
from scipy import ndimage, signal
from sklearn.decomposition import PCA
from typing import Dict, Tuple, Optional, Any
import logging
logger = logging.getLogger(__name__)
class TextureExtractor:
"""Extracts texture features from images."""
def __init__(self,
lbp_points: int = 8,
lbp_radius: int = 1,
hog_orientations: int = 9,
hog_pixels_per_cell: Tuple[int, int] = (8, 8),
hog_cells_per_block: Tuple[int, int] = (2, 2),
lacunarity_window: int = 15,
ehd_threshold: float = 0.3,
angle_resolution: int = 45):
"""
Initialize texture extractor.
Args:
lbp_points: Number of points for LBP
lbp_radius: Radius for LBP
hog_orientations: Number of orientations for HOG
hog_pixels_per_cell: Pixels per cell for HOG
hog_cells_per_block: Cells per block for HOG
lacunarity_window: Window size for lacunarity
ehd_threshold: Threshold for EHD
angle_resolution: Angle resolution for EHD
"""
self.lbp_points = lbp_points
self.lbp_radius = lbp_radius
self.hog_orientations = hog_orientations
self.hog_pixels_per_cell = hog_pixels_per_cell
self.hog_cells_per_block = hog_cells_per_block
self.lacunarity_window = lacunarity_window
self.ehd_threshold = ehd_threshold
self.angle_resolution = angle_resolution
def extract_lbp(self, gray_image: np.ndarray) -> np.ndarray:
"""
Extract Local Binary Pattern features.
Args:
gray_image: Grayscale input image
Returns:
LBP feature map
"""
try:
lbp = local_binary_pattern(
gray_image,
self.lbp_points,
self.lbp_radius,
method='uniform'
)
return self._convert_to_uint8(lbp)
except Exception as e:
logger.error(f"LBP extraction failed: {e}")
return np.zeros_like(gray_image, dtype=np.uint8)
def extract_hog(self, gray_image: np.ndarray) -> np.ndarray:
"""
Extract Histogram of Oriented Gradients features.
Args:
gray_image: Grayscale input image
Returns:
HOG feature map
"""
try:
_, vis = hog(
gray_image,
orientations=self.hog_orientations,
pixels_per_cell=self.hog_pixels_per_cell,
cells_per_block=self.hog_cells_per_block,
visualize=True,
feature_vector=True
)
return exposure.rescale_intensity(vis, out_range=(0, 255)).astype(np.uint8)
except Exception as e:
logger.error(f"HOG extraction failed: {e}")
return np.zeros_like(gray_image, dtype=np.uint8)
def compute_local_lacunarity(self, gray_image: np.ndarray, window_size: int) -> np.ndarray:
"""
Compute local lacunarity.
Args:
gray_image: Grayscale input image
window_size: Size of the sliding window
Returns:
Local lacunarity map
"""
try:
arr = gray_image.astype(np.float32)
m1 = ndimage.uniform_filter(arr, size=window_size)
m2 = ndimage.uniform_filter(arr * arr, size=window_size)
var = m2 - m1 * m1
eps = 1e-6
lac = var / (m1 * m1 + eps) + 1
lac[m1 <= eps] = 0
return lac
except Exception as e:
logger.error(f"Local lacunarity computation failed: {e}")
return np.zeros_like(gray_image, dtype=np.float32)
def compute_lacunarity_features(self, gray_image: np.ndarray) -> Tuple[np.ndarray, np.ndarray, np.ndarray]:
"""
Compute three types of lacunarity features.
Args:
gray_image: Grayscale input image
Returns:
Tuple of (lac1, lac2, lac3) lacunarity maps
"""
try:
# L1: Single window lacunarity
lac1 = self.compute_local_lacunarity(gray_image, self.lacunarity_window)
# L2: Multi-scale lacunarity
scales = [max(3, self.lacunarity_window//2), self.lacunarity_window, self.lacunarity_window*2]
lac2 = np.mean([
self.compute_local_lacunarity(gray_image, s) for s in scales
], axis=0)
# L3: DBC Lacunarity (if available)
try:
from ..models.dbc_lacunarity import DBC_Lacunarity
x = torch.from_numpy(gray_image.astype(np.float32)/255.0)[None, None]
layer = DBC_Lacunarity(window_size=self.lacunarity_window).eval()
with torch.no_grad():
lac3 = layer(x).squeeze().cpu().numpy()
except ImportError:
logger.warning("DBC Lacunarity not available, using L2 as L3")
lac3 = lac2.copy()
return (
self._convert_to_uint8(lac1),
self._convert_to_uint8(lac2),
self._convert_to_uint8(lac3)
)
except Exception as e:
logger.error(f"Lacunarity features computation failed: {e}")
empty = np.zeros_like(gray_image, dtype=np.uint8)
return empty, empty, empty
def generate_ehd_masks(self, mask_size: int = 3) -> np.ndarray:
"""
Generate masks for Edge Histogram Descriptor.
Args:
mask_size: Size of the mask
Returns:
Array of EHD masks
"""
if mask_size < 3:
mask_size = 3
if mask_size % 2 == 0:
mask_size += 1
# Base gradient mask
Gy = np.outer([1, 0, -1], [1, 2, 1])
# Expand if needed
if mask_size > 3:
expd = np.outer([1, 2, 1], [1, 2, 1])
for _ in range((mask_size - 3) // 2):
Gy = signal.convolve2d(expd, Gy, mode='full')
# Generate masks for different angles
angles = np.arange(0, 360, self.angle_resolution)
masks = np.zeros((len(angles), mask_size, mask_size), dtype=np.float32)
for i, angle in enumerate(angles):
masks[i] = ndimage.rotate(Gy, angle, reshape=False, mode='nearest')
return masks
def extract_ehd_features(self, gray_image: np.ndarray) -> Tuple[np.ndarray, np.ndarray]:
"""
Extract Edge Histogram Descriptor features.
Args:
gray_image: Grayscale input image
Returns:
Tuple of (ehd_features, ehd_map)
"""
try:
# Generate masks
masks = self.generate_ehd_masks()
# Convert to tensor
X = torch.from_numpy(gray_image.astype(np.float32)/255.0).unsqueeze(0).unsqueeze(0)
masks_tensor = torch.tensor(masks).unsqueeze(1).float()
# Convolve with masks
edge_responses = F.conv2d(X, masks_tensor, dilation=7)
# Find maximum response
values, indices = torch.max(edge_responses, dim=1)
indices[values < self.ehd_threshold] = masks.shape[0]
# Pool features
feat_vect = []
for edge in range(masks.shape[0] + 1):
pooled = F.avg_pool2d(
(indices == edge).unsqueeze(1).float(),
kernel_size=5, stride=1, padding=2
)
feat_vect.append(pooled.squeeze(1))
ehd_features = torch.stack(feat_vect, dim=1).squeeze(0).cpu().numpy()
ehd_map = np.argmax(ehd_features, axis=0).astype(np.uint8)
return ehd_features, ehd_map
except Exception as e:
logger.error(f"EHD features extraction failed: {e}")
empty_features = np.zeros((9, gray_image.shape[0]-4, gray_image.shape[1]-4), dtype=np.float32)
empty_map = np.zeros_like(gray_image, dtype=np.uint8)
return empty_features, empty_map
def extract_all_texture_features(self, gray_image: np.ndarray) -> Dict[str, np.ndarray]:
"""
Extract all texture features from a grayscale image.
Args:
gray_image: Grayscale input image
Returns:
Dictionary of texture features
"""
features = {}
try:
# LBP
features['lbp'] = self.extract_lbp(gray_image)
# HOG
features['hog'] = self.extract_hog(gray_image)
# Lacunarity
lac1, lac2, lac3 = self.compute_lacunarity_features(gray_image)
features['lac1'] = lac1
features['lac2'] = lac2
features['lac3'] = lac3
# EHD
ehd_features, ehd_map = self.extract_ehd_features(gray_image)
features['ehd_features'] = ehd_features
features['ehd_map'] = ehd_map
logger.debug("All texture features extracted successfully")
except Exception as e:
logger.error(f"Texture feature extraction failed: {e}")
# Return empty features
features = {
'lbp': np.zeros_like(gray_image, dtype=np.uint8),
'hog': np.zeros_like(gray_image, dtype=np.uint8),
'lac1': np.zeros_like(gray_image, dtype=np.uint8),
'lac2': np.zeros_like(gray_image, dtype=np.uint8),
'lac3': np.zeros_like(gray_image, dtype=np.uint8),
'ehd_features': np.zeros((9, gray_image.shape[0]-4, gray_image.shape[1]-4), dtype=np.float32),
'ehd_map': np.zeros_like(gray_image, dtype=np.uint8)
}
return features
def _convert_to_uint8(self, arr: np.ndarray) -> np.ndarray:
"""Convert array to uint8 with proper normalization."""
arr = np.nan_to_num(arr, nan=0.0, posinf=0.0, neginf=0.0)
if arr.ptp() > 0:
normalized = (arr - arr.min()) / (arr.ptp() + 1e-6) * 255
else:
normalized = np.zeros_like(arr)
return np.clip(normalized, 0, 255).astype(np.uint8)
def compute_texture_statistics(self, features: Dict[str, np.ndarray],
mask: Optional[np.ndarray] = None) -> Dict[str, Dict[str, float]]:
"""
Compute statistics for texture features.
Args:
features: Dictionary of texture features
mask: Optional mask to apply
Returns:
Dictionary of feature statistics
"""
stats = {}
for feature_name, feature_data in features.items():
if feature_name == 'ehd_features':
# Special handling for EHD features
if mask is not None:
# Apply mask to each channel
masked_features = []
for i in range(feature_data.shape[0]):
channel = feature_data[i]
if mask.shape != channel.shape:
# Resize mask to match channel
mask_resized = cv2.resize(mask, (channel.shape[1], channel.shape[0]),
interpolation=cv2.INTER_NEAREST)
masked_channel = np.where(mask_resized > 0, channel, np.nan)
else:
masked_channel = np.where(mask > 0, channel, np.nan)
masked_features.append(masked_channel)
feature_data = np.stack(masked_features, axis=0)
else:
feature_data = feature_data
# Compute statistics for each EHD channel
channel_stats = {}
for i in range(feature_data.shape[0]):
channel = feature_data[i]
valid_data = channel[~np.isnan(channel)]
if len(valid_data) > 0:
channel_stats[f'channel_{i}'] = {
'mean': float(np.mean(valid_data)),
'std': float(np.std(valid_data)),
'min': float(np.min(valid_data)),
'max': float(np.max(valid_data)),
'median': float(np.median(valid_data))
}
stats[feature_name] = channel_stats
else:
# Regular 2D features
if mask is not None and mask.shape == feature_data.shape:
masked_data = np.where(mask > 0, feature_data, np.nan)
else:
masked_data = feature_data
valid_data = masked_data[~np.isnan(masked_data)]
if len(valid_data) > 0:
stats[feature_name] = {
'mean': float(np.mean(valid_data)),
'std': float(np.std(valid_data)),
'min': float(np.min(valid_data)),
'max': float(np.max(valid_data)),
'median': float(np.median(valid_data))
}
else:
stats[feature_name] = {
'mean': 0.0, 'std': 0.0, 'min': 0.0, 'max': 0.0, 'median': 0.0
}
return stats
|