Segregates images using feature extraction and Clustering algorithm

• About the Project
  • Aim
  • Description
  • Tech Stack
  • File Structure
• Getting Started
  • Prerequisites
  • Installation
• Usage
• Theory and Approach
• Results and Demo
• Future Work
• Troubleshooting
• Contributors
• Acknowledgements and Resources
• License

This project aims at creating an image grouping algorithm. The algorithm should be able to group similar images on the basis of extracted features.

We have used ORB algorithm for extracting features and Scikit K-means clustering algorithm to clusterize images. So it reads images from a folder and applies ORB to all images to give its descriptors and finds optimum no. of groups(K) then applies K-means on descriptors and paste images to their respective cluster folder. For more info refer our report

• OpenCV
• Scikit Learn

.📦
├── 📂assets				# contains images and video			
│   ├── 📜demo.mp4												 
│   └── 📜overview.png									  
├── 📜kmeansUsingOpenCV.py		# code with openCV kmeans 
├── 📜kmeansUsingSkLearn.py		# code with SkLearn kmeans
├── 📜LICENSE				# MIT license
├── 📜README.md				
└── 📂report				# Project report
    └── 📜report.pdf

• Should have python environment. You can refer here for the setup.
• Python librairies
  • OpenCV pip install opencv-python
  • Scikit pip install scikit-learn
  • numpy pip install numpy
  • kneed pip install kneed

For installation of pip you can refer here

• Must have Test Data, or you can get it from here and from kaggle.

1. Clone the repo

   git clone https://github.com/shahpratham/Image_Grouping.git
   

cd /path/to/Image_Grouping

For using Sklearn K-means use this, similar can be done openCV code

python kmeansUsingSkLearn.py

For viewing Cluster Directory

cd /path/to/TestDAta

Select assorted images of single label test subjects like for example cats & cars. Apply the clustering algorithm to find images of cats in one folder & cars in a seperate folder. It reads images from a folder and applies ORB to all images to give its descriptors and find optimum value of k and applies K-means on descriptors and paste images to their respective cluster folder.

• Preprocessing

To group images, it requires processing the images under test. Image processing is the operation of converting images into computer readable data. To perform the necessary operations we processed the images using the OpenCV python library which allows us to read images in matrix format.

• Feature extraction via Orb

Now we need to extract features from the image to understand the contents of the image. A feature / keypoint is a piece of information about the content of an image; typically about whether a certain region of the image has certain properties.These features are stored in the computer memory in the form of descriptors. The descriptor contains the visual description of the patch and is used to compare the similarity between image features. So, by applying openCV ORB to all images, we stored all keypoints and descriptors of images in the list.

• K-Means clustering

So, after getting descriptors of all images, we need to cluster them by using K-Means clustering. First we need to find no. of clusters so we are doing that by applying the elbow method using distortions. K-means clustering is a method of vector quantization, originally from signal processing, that aims to partition n observations into k clusters in which each observation belongs to the cluster with the nearest mean.

• Making directories for cluster and pasting image to its respective directory(cluster)

Now we have all images labeled(cluster number), we can group them by making separate directories using os library and copy paste images from the main folder to their respective directories by shutil.

Did this for 101 images and tested on Lenovo Legion 5(Ryzen 5 4600H) and got accuracy close to 80%. We tried on our K-Means code from scratch(you can get in dev branch-- kmeans.py) and got accuracy around 55-65%.

• Finding optimum of K
• Improving accuracy
• Make a web app where one can upload images and can cluster it online.

• Changing parameters like nfeatures,  edgeThresholdand fastThreshold in ORB_create can prove to be effective for some datasets
• The ideal way for dealing with outliers was to increase your test data
• After changing max cluster value from 10 to 20, KneeLocator was able to find optimum value

• Pratham Shah
• Yash Deshpande

• SRA VJTI Eklavya 2021
• Our mentors Mann Doshi and Prathamesh Tagore for their guidance throughout this project
• K-means Research Paper
• ORB Research Paper
• Optimum K
• For more resources refer References section in report

The License used in this project