Image Features Extraction with Machine Learning

A local image characteristic is a tiny patch in the image that is indifferent to the image scaling, rotation, and lighting change. It’s like the tip of a tower or the corner of a window in the image below. In this article, I will walk you through the task of image features extraction with Machine Learning.

Unlike a random point on the background of the image above, the tip of the tower can be accurately detected in most images of the same scene. It is geometric and photometrically invariant.

Also, Read – A/B Testing in Machine Learning.

Characteristics of Good Features:

A good local feature is like the piece you start with when solving a puzzle, except on a much smaller scale. It’s the cat’s eye or the corner of the table, not a piece on a white wall.

The extracted image features must be:

  • Repeatable and precise so that they can be extracted from different images showing the same object.
  • Distinctive to the image, so images with different structure will not have them.

Image Features Extraction

Let’s start with importing the necessary libraries for the task of image features extraction:

import numpy as np import pandas as pd import cv2 import matplotlib.pyplot as plt import os

Now, I will load the image that we need for this task:

img_building = cv2.imread(os.path.join('building_1.jpg')) img_building = cv2.cvtColor(img_building, cv2.COLOR_BGR2RGB) orb = cv2.ORB_create() # OpenCV 3 backward incompatibility: Do not create a detector with `cv2.ORB()`. key_points, description = orb.detectAndCompute(img_building, None) img_building_keypoints = cv2.drawKeypoints(img_building, key_points, img_building, flags=cv2.DRAW_MATCHES_FLAGS_DRAW_RICH_KEYPOINTS) # Draw circles. plt.figure(figsize=(16, 16)) plt.title('ORB Interest Points') plt.imshow(img_building_keypoints)

The points of image features found are circled in the image above. As we can see, some of these points are unique to this building, like the points near the top of the two towers. However, others like those at the top of the tree may not be distinctive.

Feature Matching

Now, let’s see if we can extract the same features from the image that is taken from a different angle:

def image_detect_and_compute(detector, img_name): img = cv2.imread(img_name)) img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY) kp, des = detector.detectAndCompute(img, None) return img, kp, des def draw_image_matches(detector, img1_name, img2_name, nmatches=10): img1, kp1, des1 = image_detect_and_compute(detector, img1_name) img2, kp2, des2 = image_detect_and_compute(detector, img2_name) bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=True) matches = bf.match(des1, des2) matches = sorted(matches, key = lambda x: x.distance) # Sort matches by distance. Best come first. img_matches = cv2.drawMatches(img1, kp1, img2, kp2, matches[:nmatches], img2, flags=2) # Show top 10 matches plt.figure(figsize=(16, 16)) plt.title(type(detector)) plt.imshow(img_matches) orb = cv2.ORB_create() draw_image_matches(orb, 'building_1.jpg', 'building_2.jpg')
sift = cv2.xfeatures2d.SIFT_create() kp, des = sift.detectAndCompute(img_building, None) img_kp = cv2.drawKeypoints(img_building, kp, img_building) plt.figure(figsize=(15, 15)) plt.imshow(img_kp)

Also, Read – Lambda Expression in Python.

More than half of the first 10 matches were correct. In real applications, instead of using the first n matches, a match distance threshold is used to filter out spurious matches. I hope you liked this article on how we can extract image features using Machine Learning. Feel free to ask your valuable questions in the comments section below.

Follow Us:

Default image
Aman Kharwal

I am a programmer from India, and I am here to guide you with Data Science, Machine Learning, Python, and C++ for free. I hope you will learn a lot in your journey towards Coding, Machine Learning and Artificial Intelligence with me.

Leave a Reply