Similar Image Finder

This model will find the n similar images from the training dataset of the input image. It leverages the feature extraction of pre-trained of resnet50 and KNN algorithm to find similar images.

Libraries:

Keras,sklearn,skimage

Workflow:

At first, pretrained resnet50 will be used to extract features of images.We downloaded only the feature extraction layer of resnet50 and paased all training images through it.
After passing it from resnet50 , we have flattened it and fitted it through knn where we used cosine distance and number of neighbours will depend on user input.
test image will go to resnet50 and then it will get nearest neighbours from the previously fitted knn and the similar image will be stored in output folder.

Instruction to run:

Clone this repo. 2.Check if all necessary libraries are present or not
Navigate to the project folder. put training dataset in data/train folder(only .jpeg and .jpg supported now) and your query images in test folder.
Invoke the main program file by running: python similar_image.py and when prompted provide the number of similar image you want to be fetched. 5.After completion check the newly created output folder for results.

folder strucutre:

Similar_image.py: It containes the core code. It needs to be invoked to get the similar images. It will ask for how many similar image to be fetched as input.

data: In this folder training data will reside in train folder. Query image will reside in test folder(you can keep as many as you want).

output: this folder will automatically be created results.

Other files contains various utility functions that will be used in core program file.

Results:

Here are some results: