The Power of HOG in Object Detection and Recognition with Amazon SageMaker

Introduction to HOG

The HOG (Histogram of Oriented Gradients) feature descriptor is a popular technique for object detection and recognition in computer vision.

It works by dividing an image into small cells, computing the gradient orientation of each pixel within each cell, and then creating a histogram of these orientations for each block of cells.

When using the HOG descriptor for image similarity, one approach is to directly compare the histograms of two images using a distance metric such as the Euclidean distance or cosine similarity. However, this approach can be sensitive to changes in lighting, contrast, and other factors that affect the overall intensity of the image.

To address this issue, joint descriptors can capture more robust and discriminative information about image features. A joint descriptor combines multiple features, such as color, texture, and shape, into a single vector representation. This allows for more comprehensive image comparisons that consider multiple aspects of image content.

One example of a joint descriptor that can be used with HOG is the Histogram of Oriented Gradients and Color (HOG+C) descriptor. This descriptor combines the HOG feature descriptor with color information by computing each cell’s histogram of color values and the gradient orientations. This can improve the robustness of the descriptor to changes in lighting and contrast while still capturing important information about edge and texture features.

To compare images using a joint descriptor like HOG+C, a distance metric such as the cosine similarity can be used to measure the similarity between the vector representations of two images. This can be useful for tasks such as image retrieval, where the goal is to find images similar to a given query image based on their visual content.

Amazon SageMaker

Amazon SageMaker is a managed machine learning service provided by Amazon Web Services (AWS) that enables data scientists and developers to build, train, and deploy machine learning models at scale. Amazon SageMaker provides a fully managed environment to run your machine learning algorithms with built-in algorithms for common machine learning problems, including object detection.

Pioneers in Cloud Consulting & Migration Services

Reduced infrastructural costs
Accelerated application deployment

Get Started

Step-by-Step Guide

To use HOG descriptors in Amazon SageMaker, we need to create a notebook instance on Amazon SageMaker. This instance will be used to run our HOG feature extraction code. Once the instance is created, we can launch the Jupyter Notebook environment and start working on our code.
Load the images that we want to extract HOG features from.
Convert them to grayscale. This is because HOG works on the gradient of the image, which is more informative in grayscale.
Using the HOGDescriptor() function. We will also specify the number of cells per block and the number of bins in the histogram.
The setSVMDetector() function sets the HOG descriptor to use the default SVM detector, trained to detect people. We can also train our SVM detector using a labeled image dataset.
Next, we need to compute the HOG descriptors for each cell in the images. This is done using the compute() function of the HOG descriptor object.
The compute() function returns a 1D array of HOG features for the entire image. We can reshape this array into a 2D array representing each cell’s HOG features.
Now that we have the HOG features for each image, we can use cosine similarity to compare and detect similarities.
To compute cosine similarity, we first need to normalize the HOG features for each image using the L2-norm. This is done using the normalize() function from NumPy.

import cv2
import numpy as np
import matplotlib.pyplot as plt
car_img = cv2.imread('car.jpg')
person_img = cv2.imread('person.jpg')
car_gray = cv2.cvtColor(car_img, cv2.COLOR_BGR2GRAY)
person_gray = cv2.cvtColor(person_img, cv2.COLOR_BGR2GRAY)
hog = cv2.HOGDescriptor()
hog.setSVMDetector(cv2.HOGDescriptor_getDefaultPeopleDetector())
car_hist = hog.compute(car_gray)
person_hist = hog.compute(person_gray)
car_hist = car_hist.reshape(-1, hog.getDescriptorSize())
person_hist = person_hist.reshape(-1, hog.getDescriptorSize())
car_norm = np.linalg.norm(car_hist, axis=1)
person_norm = np.linalg.norm(person_hist, axis=1)
car_hist_norm = car_hist / car_norm.reshape(-1, 1)
person_hist_norm = person_hist / person_norm.reshape(-1, 1)

import cv2

import numpy as np

import matplotlib.pyplot as plt

car_img = cv2.imread('car.jpg')

person_img = cv2.imread('person.jpg')

car_gray = cv2.cvtColor(car_img, cv2.COLOR_BGR2GRAY)

person_gray = cv2.cvtColor(person_img, cv2.COLOR_BGR2GRAY)

hog = cv2.HOGDescriptor()

hog.setSVMDetector(cv2.HOGDescriptor_getDefaultPeopleDetector())

car_hist = hog.compute(car_gray)

person_hist = hog.compute(person_gray)

car_hist = car_hist.reshape(-1, hog.getDescriptorSize())

person_hist = person_hist.reshape(-1, hog.getDescriptorSize())

car_norm = np.linalg.norm(car_hist, axis=1)

person_norm = np.linalg.norm(person_hist, axis=1)

car_hist_norm = car_hist / car_norm.reshape(-1, 1)

person_hist_norm = person_hist / person_norm.reshape(-1, 1)

Conclusion

Histogram of Oriented Gradients (HOG) feature descriptor is a powerful tool for extracting useful information from images for computer vision tasks such as object detection and recognition. Its ability to capture edge and shape information while being computationally efficient makes it a popular choice for many applications. However, it is important to note that HOG may not perform as well in all scenarios. It should be optimized with proper parameter tuning and other techniques, such as data augmentation, to achieve the best performance. Overall, HOG remains a valuable and widely used feature descriptor in computer vision.

Empowering organizations to become ‘data driven’ enterprises with our Cloud experts.

Reduced infrastructure costs
Timely data-driven decisions

Get Started

About CloudThat

CloudThat is an official AWS (Amazon Web Services) Advanced Consulting Partner and Training partner and Microsoft Gold Partner, helping people develop knowledge of the cloud and help their businesses aim for higher goals using best-in-industry cloud computing practices and expertise. We are on a mission to build a robust cloud computing ecosystem by disseminating knowledge on technological intricacies within the cloud space. Our blogs, webinars, case studies, and white papers enable all the stakeholders in the cloud computing sphere.

Drop a query if you have any questions regarding HOG, I will get back to you quickly.

To get started, go through our Consultancy page and Managed Services Package, CloudThat’s offerings.

FAQs

1. What are some applications of the HOG feature descriptor?

ANS: – HOG is commonly used in object detection and recognition tasks, such as pedestrian detection, face detection, and vehicle detection. It is also used in other computer vision applications, such as image segmentation and optical character recognition (OCR).

2. How does the HOG feature descriptor compare to other feature extraction methods?

ANS: – HOG is often compared to other feature extraction methods, such as SIFT and SURF. While these methods can also be effective, HOG has some advantages, such as being computationally efficient and able to handle large variations in object appearance. However, HOG may not perform as well in cases where there are complex backgrounds or occlusions.

3. How can the HOG feature descriptor be optimized for better performance?

ANS: – HOG performance can be optimized by adjusting the parameters used in the feature extraction process, such as the cell size, block size, and number of histogram bins. These parameters can be tuned using cross-validation techniques on a training dataset to maximize performance on a validation dataset. Other techniques, such as data augmentation, image normalization, and multiple scales, can improve HOG performance.

WRITTEN BY Hridya Hari

Hridya Hari works as a Research Associate - Data and AIoT at CloudThat. She is a data science aspirant who is also passionate about cloud technologies. Her expertise also includes Exploratory Data Analysis.