NiranjanSathish's picture
Upload 41 files
f647a80 verified
import numpy as np
import cv2
from sklearn.cluster import KMeans
from sklearn.preprocessing import LabelEncoder
import matplotlib.pyplot as plt
from typing import Tuple, List, Dict, Optional
from dataclasses import dataclass
from enum import Enum
import seaborn as sns
class ColorClassificationMethod(Enum):
"""Different methods for classifying cell colors."""
DOMINANT_COLOR = "dominant_color"
AVERAGE_COLOR = "average_color"
HISTOGRAM_BINS = "histogram_bins"
HSV_QUANTIZATION = "hsv_quantization"
@dataclass
class GridCell:
"""Represents a single grid cell with its properties."""
row: int
col: int
average_color: np.ndarray
dominant_color: np.ndarray
brightness: float
saturation: float
hue: float
color_category: int
pixel_data: np.ndarray
class ImageGridAnalyzer:
"""
Analyzes images by dividing them into grids and classifying each cell's color properties.
Uses vectorized NumPy operations for high performance.
"""
def __init__(self, grid_size: Tuple[int, int] = (32, 32),
classification_method: ColorClassificationMethod = ColorClassificationMethod.DOMINANT_COLOR,
n_color_categories: int = 16):
"""
Initialize the grid analyzer.
Args:
grid_size: (rows, cols) for the grid division
classification_method: Method to classify cell colors
n_color_categories: Number of color categories for classification
"""
self.grid_size = grid_size
self.classification_method = classification_method
self.n_color_categories = n_color_categories
self.color_classifier = None
self.category_colors = None
def divide_image_into_grid(self, image: np.ndarray) -> Tuple[np.ndarray, Tuple[int, int]]:
"""
Divide image into a grid using vectorized operations.
Args:
image: Input image (H, W, C)
Returns:
Grid of cells (grid_rows, grid_cols, tile_height, tile_width, channels)
Tuple of (tile_height, tile_width)
"""
h, w, c = image.shape
grid_rows, grid_cols = self.grid_size
# Calculate tile dimensions
tile_h = h // grid_rows
tile_w = w // grid_cols
# Adjust image size to fit grid perfectly (crop if necessary)
adjusted_h = tile_h * grid_rows
adjusted_w = tile_w * grid_cols
image = image[:adjusted_h, :adjusted_w]
# Vectorized grid division using reshape and transpose
# This is much faster than nested loops
grid = image.reshape(grid_rows, tile_h, grid_cols, tile_w, c)
grid = grid.transpose(0, 2, 1, 3, 4) # (grid_rows, grid_cols, tile_h, tile_w, c)
return grid, (tile_h, tile_w)
def analyze_grid_colors_vectorized(self, grid: np.ndarray) -> Dict[str, np.ndarray]:
"""
Analyze color properties of all grid cells using vectorized operations.
Args:
grid: Grid of cells (grid_rows, grid_cols, tile_h, tile_w, c)
Returns:
Dictionary containing vectorized analysis results
"""
grid_rows, grid_cols, tile_h, tile_w, c = grid.shape
# Reshape for vectorized operations: (total_cells, pixels_per_cell, channels)
cells_flat = grid.reshape(grid_rows * grid_cols, tile_h * tile_w, c)
# Calculate average colors for all cells at once
average_colors = np.mean(cells_flat, axis=1) # (total_cells, c)
# Calculate dominant colors using vectorized approach
dominant_colors = self._calculate_dominant_colors_vectorized(cells_flat)
# Convert to HSV for additional analysis
hsv_averages = self._rgb_to_hsv_vectorized(average_colors)
# Calculate brightness (V in HSV)
brightness = hsv_averages[:, 2]
# Calculate saturation
saturation = hsv_averages[:, 1]
# Calculate hue
hue = hsv_averages[:, 0]
# Reshape results back to grid format
results = {
'average_colors': average_colors.reshape(grid_rows, grid_cols, c),
'dominant_colors': dominant_colors.reshape(grid_rows, grid_cols, c),
'brightness': brightness.reshape(grid_rows, grid_cols),
'saturation': saturation.reshape(grid_rows, grid_cols),
'hue': hue.reshape(grid_rows, grid_cols),
'cells_data': grid # Keep original cell data
}
return results
def _calculate_dominant_colors_vectorized(self, cells_flat: np.ndarray) -> np.ndarray:
"""
Calculate dominant color for each cell using vectorized operations.
Args:
cells_flat: Flattened cells (total_cells, pixels_per_cell, channels)
Returns:
Dominant colors for all cells (total_cells, channels)
"""
import warnings
total_cells, pixels_per_cell, c = cells_flat.shape
dominant_colors = np.zeros((total_cells, c))
# Process cells in batches for memory efficiency
batch_size = 100
for i in range(0, total_cells, batch_size):
end_idx = min(i + batch_size, total_cells)
batch = cells_flat[i:end_idx]
for j, cell_pixels in enumerate(batch):
# Check for color diversity first
unique_pixels = np.unique(cell_pixels, axis=0)
if len(unique_pixels) >= 3 and pixels_per_cell > 100:
# Use k-means for larger cells with sufficient color diversity
with warnings.catch_warnings():
warnings.filterwarnings("ignore", category=UserWarning)
warnings.filterwarnings("ignore", message=".*ConvergenceWarning.*")
kmeans = KMeans(n_clusters=min(3, len(unique_pixels)),
random_state=42, n_init=5)
labels = kmeans.fit_predict(cell_pixels)
# Get the most frequent cluster center
unique_labels, counts = np.unique(labels, return_counts=True)
dominant_idx = unique_labels[np.argmax(counts)]
dominant_colors[i + j] = kmeans.cluster_centers_[dominant_idx]
elif len(unique_pixels) >= 2:
# Use most frequent color for limited diversity
unique_colors, counts = np.unique(cell_pixels, axis=0, return_counts=True)
dominant_colors[i + j] = unique_colors[np.argmax(counts)]
else:
# Use simple average for uniform cells
dominant_colors[i + j] = np.mean(cell_pixels, axis=0)
return dominant_colors
def _rgb_to_hsv_vectorized(self, rgb_colors: np.ndarray) -> np.ndarray:
"""
Convert RGB colors to HSV using vectorized operations.
Args:
rgb_colors: RGB colors (N, 3)
Returns:
HSV colors (N, 3)
"""
# Normalize to 0-1 range
rgb_normalized = rgb_colors / 255.0
# Create a dummy image for cv2 conversion
dummy_img = rgb_normalized.reshape(-1, 1, 3).astype(np.float32)
hsv_img = cv2.cvtColor(dummy_img, cv2.COLOR_RGB2HSV)
hsv_colors = hsv_img.reshape(-1, 3)
return hsv_colors
def classify_colors(self, color_data: Dict[str, np.ndarray]) -> np.ndarray:
"""
Classify each grid cell into color categories.
Args:
color_data: Dictionary containing color analysis results
Returns:
Color categories for each grid cell (grid_rows, grid_cols)
"""
import warnings
if self.classification_method == ColorClassificationMethod.AVERAGE_COLOR:
features = color_data['average_colors']
elif self.classification_method == ColorClassificationMethod.DOMINANT_COLOR:
features = color_data['dominant_colors']
elif self.classification_method == ColorClassificationMethod.HSV_QUANTIZATION:
# Combine HSV features
h = color_data['hue']
s = color_data['saturation']
v = color_data['brightness']
features = np.stack([h, s, v], axis=-1)
else:
features = color_data['average_colors']
# Flatten for clustering
grid_rows, grid_cols = features.shape[:2]
features_flat = features.reshape(-1, features.shape[-1])
# Check for sufficient diversity before clustering
unique_features = np.unique(features_flat, axis=0)
effective_clusters = min(self.n_color_categories, len(unique_features))
if effective_clusters < 2:
# Handle case with very limited color diversity
print(f"Warning: Only {len(unique_features)} unique colors found. Using simple classification.")
categories = np.zeros(len(features_flat), dtype=int)
categories_grid = categories.reshape(grid_rows, grid_cols)
self.category_colors = unique_features[:1] if len(unique_features) > 0 else np.array([[128, 128, 128]])
return categories_grid
# Fit color classifier with warning suppression
with warnings.catch_warnings():
warnings.filterwarnings("ignore", category=UserWarning)
warnings.filterwarnings("ignore", message=".*ConvergenceWarning.*")
self.color_classifier = KMeans(n_clusters=effective_clusters,
random_state=42, n_init=10)
categories = self.color_classifier.fit_predict(features_flat)
# Store category representative colors
self.category_colors = self.color_classifier.cluster_centers_
# Reshape back to grid
categories_grid = categories.reshape(grid_rows, grid_cols)
return categories_grid
def apply_thresholding(self, color_data: Dict[str, np.ndarray],
brightness_threshold: float = 0.5,
saturation_threshold: float = 0.3) -> Dict[str, np.ndarray]:
"""
Apply thresholding to create binary masks for different criteria.
Args:
color_data: Color analysis results
brightness_threshold: Threshold for bright/dark classification
saturation_threshold: Threshold for saturated/desaturated classification
Returns:
Dictionary containing various threshold masks
"""
brightness = color_data['brightness']
saturation = color_data['saturation']
# Normalize brightness and saturation to 0-1 range
brightness_norm = brightness / 255.0 if brightness.max() > 1.0 else brightness
saturation_norm = saturation / 255.0 if saturation.max() > 1.0 else saturation
thresholds = {
'bright_mask': brightness_norm > brightness_threshold,
'dark_mask': brightness_norm <= brightness_threshold,
'saturated_mask': saturation_norm > saturation_threshold,
'desaturated_mask': saturation_norm <= saturation_threshold,
'bright_saturated': (brightness_norm > brightness_threshold) &
(saturation_norm > saturation_threshold),
'dark_saturated': (brightness_norm <= brightness_threshold) &
(saturation_norm > saturation_threshold)
}
return thresholds
def analyze_image_complete(self, image: np.ndarray) -> Dict:
"""
Complete analysis pipeline for an image.
Args:
image: Input image (H, W, C)
Returns:
Complete analysis results
"""
print(f"Analyzing image with {self.grid_size[0]}x{self.grid_size[1]} grid...")
# Step 1: Divide into grid
grid, tile_size = self.divide_image_into_grid(image)
print(f"Created grid with tile size: {tile_size}")
# Step 2: Analyze colors (vectorized)
color_data = self.analyze_grid_colors_vectorized(grid)
print("Completed color analysis")
# Step 3: Classify colors
color_categories = self.classify_colors(color_data)
print(f"Classified into {self.n_color_categories} color categories")
# Step 4: Apply thresholding
thresholds = self.apply_thresholding(color_data)
print("Applied thresholding")
# Combine all results
results = {
'grid': grid,
'tile_size': tile_size,
'color_data': color_data,
'color_categories': color_categories,
'thresholds': thresholds,
'category_colors': self.category_colors
}
return results
def visualize_analysis(self, results: Dict, original_image: np.ndarray):
"""
Create comprehensive visualizations of the analysis results.
Args:
results: Analysis results from analyze_image_complete
original_image: Original input image
"""
fig, axes = plt.subplots(2, 4, figsize=(20, 10))
# Original image
axes[0, 0].imshow(original_image)
axes[0, 0].set_title('Original Image')
axes[0, 0].axis('off')
# Average colors
avg_colors = results['color_data']['average_colors'].astype(np.uint8)
axes[0, 1].imshow(avg_colors)
axes[0, 1].set_title('Average Colors per Cell')
axes[0, 1].axis('off')
# Dominant colors
dom_colors = results['color_data']['dominant_colors'].astype(np.uint8)
axes[0, 2].imshow(dom_colors)
axes[0, 2].set_title('Dominant Colors per Cell')
axes[0, 2].axis('off')
# Color categories
categories = results['color_categories']
im_cat = axes[0, 3].imshow(categories, cmap='tab20')
axes[0, 3].set_title(f'Color Categories ({self.n_color_categories} classes)')
axes[0, 3].axis('off')
plt.colorbar(im_cat, ax=axes[0, 3])
# Brightness
brightness = results['color_data']['brightness']
im_bright = axes[1, 0].imshow(brightness, cmap='gray')
axes[1, 0].set_title('Brightness Values')
axes[1, 0].axis('off')
plt.colorbar(im_bright, ax=axes[1, 0])
# Saturation
saturation = results['color_data']['saturation']
im_sat = axes[1, 1].imshow(saturation, cmap='viridis')
axes[1, 1].set_title('Saturation Values')
axes[1, 1].axis('off')
plt.colorbar(im_sat, ax=axes[1, 1])
# Threshold: Bright areas
axes[1, 2].imshow(results['thresholds']['bright_mask'], cmap='gray')
axes[1, 2].set_title('Bright Areas (Threshold)')
axes[1, 2].axis('off')
# Threshold: Saturated areas
axes[1, 3].imshow(results['thresholds']['saturated_mask'], cmap='gray')
axes[1, 3].set_title('Saturated Areas (Threshold)')
axes[1, 3].axis('off')
plt.tight_layout()
plt.show()
# Show color category palette
self._visualize_color_palette(results['category_colors'])
def _visualize_color_palette(self, category_colors: np.ndarray):
"""
Visualize the color category palette.
Args:
category_colors: Color palette (n_categories, channels)
"""
if category_colors is None:
return
fig, ax = plt.subplots(1, 1, figsize=(12, 2))
# Normalize colors if needed
colors = category_colors.copy()
if colors.max() > 1.0:
colors = colors / 255.0
# Create color swatches
palette = colors.reshape(1, -1, 3)
ax.imshow(palette, aspect='auto')
ax.set_xlim(0, len(colors))
ax.set_ylim(0, 1)
ax.set_xticks(range(len(colors)))
ax.set_xticklabels([f'Cat {i}' for i in range(len(colors))])
ax.set_title(f'Color Category Palette ({len(colors)} categories)')
ax.set_ylabel('Color Categories')
plt.tight_layout()
plt.show()
def get_performance_stats(self, results: Dict) -> Dict:
"""
Calculate performance and analysis statistics.
Args:
results: Analysis results
Returns:
Dictionary containing statistics
"""
grid_shape = results['color_categories'].shape
total_cells = np.prod(grid_shape)
# Color diversity
unique_categories = len(np.unique(results['color_categories']))
# Brightness statistics
brightness = results['color_data']['brightness']
# Saturation statistics
saturation = results['color_data']['saturation']
stats = {
'grid_size': f"{grid_shape[0]}x{grid_shape[1]}",
'total_cells': total_cells,
'unique_color_categories': unique_categories,
'category_utilization': unique_categories / self.n_color_categories,
'avg_brightness': np.mean(brightness),
'brightness_std': np.std(brightness),
'avg_saturation': np.mean(saturation),
'saturation_std': np.std(saturation),
'bright_cells_percent': np.mean(results['thresholds']['bright_mask']) * 100,
'saturated_cells_percent': np.mean(results['thresholds']['saturated_mask']) * 100
}
return stats
# Example usage and testing
def main():
"""
Example usage of the ImageGridAnalyzer.
"""
# Create sample test image (or load your own)
def create_test_image():
# Create a colorful test image with gradients and patterns
img = np.zeros((256, 256, 3), dtype=np.uint8)
# Add gradients
for i in range(256):
for j in range(256):
img[i, j, 0] = i # Red gradient
img[i, j, 1] = j # Green gradient
img[i, j, 2] = (i + j) % 255 # Blue pattern
return img
# Initialize analyzer
analyzer = ImageGridAnalyzer(
grid_size=(32, 32), # 32x32 grid = 1024 cells
classification_method=ColorClassificationMethod.DOMINANT_COLOR,
n_color_categories=16
)
# Create or load test image
# test_image = create_test_image()
# Or load real image (uncomment and modify path):
test_image = cv2.imread('processed_quantized.jpg')
test_image = cv2.cvtColor(test_image, cv2.COLOR_BGR2RGB)
print("Starting analysis...")
# Analyze the image
results = analyzer.analyze_image_complete(test_image)
# Get performance statistics
stats = analyzer.get_performance_stats(results)
print("\n=== Analysis Statistics ===")
for key, value in stats.items():
print(f"{key}: {value}")
# Visualize results
# analyzer.visualize_analysis(results, test_image)
return results, analyzer
if __name__ == "__main__":
results, analyzer = main()