Search_engine-using-elasticsearch

Image and text search_engine using elasticsearch and and Approximate nearest neighbors.

Using text as input to our search_engine :

Using images as input to our search_engine :

About the project

This academic project represents an image and text search engine that covers a 7200 image instance of OpenImages V6 which is a large scale dataset. We performed feature extraction from the images with VGG16 model to get a feature vector of length 4096 and then we reduced these dimentions to only 512 using a PCA-Principal Component Analysis of 512 components.

Prerequisites

You must have the following Prerequisites:

• ElasticSearch 7.14.1
• Elastiknn plugin 7.14.1.0
• kibana7.14.1.0
• Python 3.9.6

Pipeline :

Our project was composed of 3 main steps :

Data preprocessing :

We performed feature extraction from the images with VGG16 model to get a feature vector of length 4096 and then we reduced these dimentions to only 512 using a PCA-Principal Component Analysis of 512 components.

Data indexing to Elasticsearch:

The feature vectors of length 512 were all indexed to an Elasticsearch Index using ElasticSearch python client that uses Elastiknn plugin. We also indexed textual data in the same index.

Application development:

To visualise how our model performed with image and text search a web application is developed using:

• ElasticSearch python client: to query elasticsearch index.
• Streamlit : to design a frontend for the application where users perform text and image search.

Contribution:

Sirine Arfa & Arij Flihi

INDP3 AIM , SUPCOM 2022/2023