- Image Clustering using simple KMeans does'nt yield good results. Simply flattening the image and passing it to KMeans doesn't preseve image features.
- Instead, Convolutional Neural Networks preserves important characteristics of an image, as its biologically inspired by the architecture that is present in human/animal brains.
- Convolutional Neural Network layers detects pixels, edges, text, parts, objects in the image, thereby preserving all the important features of an image.
- Core idea is instead of building a Convolutional Neural Network from scratch to solve our task, what if we can reuse existing trained models like VGG16, AlexNet architectures.
- Keras actually has VGG16 trained on ImageNet dataset, which is the one of the largest object classification dataset.
- Input Dataset -> My Google Photos captured on my mobile cam
- Output -> Cluster of similar Images
- Keras Model of the VGG16 network, trained on Imagenet dataset is used to extract feature vectors of the images.
- Optimal K for KMeans is determined using Elbow method
- These feature vectors of images are used to form clusters using KMeans++
- Python
Programming language - Opencv (cv2)
TO read and write the images - Numpy
To perform operations on image arrays - Keras
To load the pretrained Imagenet model weights and extract image feature vectors - Matplotlib
To plot the image clusters as a graph - os
To read and write the files from directories
