How to detect a Christmas Tree?

Using traditional strategies in image processing

CNNs are quite potent, but sometimes all we need are traditional image processing methods, especially when we can't afford extensive computational resources or pre-existing models aren't up for grabs.

Utilizing color and brightness thresholds

We will operate in the HSV color space to separate the typical colors of a Christmas Tree. We will combine this with brightness thresholds to capture the tantalizing brilliance of Christmas lights:

import cv2
import numpy as np

def apply_thresholds(image):
    # Convert image to HSV (Because Christmas Trees = Green + Bright Lights!)
    hsv = cv2.cvtColor(image, cv2.COLOR_BGR2HSV)
    
    # Define color spectrum for Christmas Trees (The "greenness" level)
    lower_green = np.array([25, 52, 72])
    upper_green = np.array([102, 255, 255])

    # Create masks based on our smart thresholds
    color_mask = cv2.inRange(hsv, lower_green, upper_green)
    brightness_mask = cv2.inRange(hsv, (0, 0, 120), (255, 255, 255))

    # Combine masks (Like combining gifts under the tree!)
    combined_mask = cv2.bitwise_or(color_mask, brightness_mask)
    return combined_mask


Engaging with shape and cluster analysis

We will employ the DBSCAN clustering algorithm to differentiate between distinct objects and their shapes. Following this, we will compute the convex hull to demarcate the borders of the detected tree:

from sklearn.cluster import DBSCAN
from scipy.spatial import ConvexHull
import matplotlib.pyplot as plt

def detect_tree_shape(image, mask):
    # Find contours - the curves joining continuous points along the borders
    contours, _ = cv2.findContours(mask, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)

    # Keeping things tight and together (just like Christmas decorations) with DBScan!
    points = np.vstack(contours).squeeze()
    clustering = DBSCAN(eps=image.diagonal().size * 0.02).fit(points)
    labels = clustering.labels_

    # Time to plot the points along with their convex hull (It's an art masterpiece!)
    unique_labels = set(labels)
    for k in unique_labels:
        if k != -1:  # Removing noise points (Because we don't like photo-bombers)
            class_member_mask = (labels == k)
            hull_points = points[class_member_mask]
            hull = ConvexHull(hull_points)
            plt.plot(hull_points[:, 0], hull_points[:, 1], 'o')
            for simplex in hull.simplices:
                plt.plot(hull_points[simplex, 0], hull_points[simplex, 1], 'k-')
    plt.show()


Deploying texture and feature selection

To differentiate the distinctive foliage of a Christmas Tree from other green entities, we explore the texture with Local Binary Patterns (LBP) and pick up the features that maximize the difference between the object and the background:

from skimage.feature import local_binary_pattern

def compute_LBP(image):
    # Convert to grayscale (Christmas trees are grayscale-friendly)
    gray_image = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)

    # Compute LBP features (Imagine you're the Sherlock Holmes of Computer Vision)
    lbp_image = local_binary_pattern(gray_image, P=8, R=1)

    # Further processing to select relevant features and spot the Christmas tree
    ...
