Image Search Engine based on it's content

Inspiration

In today era, billions of consumers of smartphones and digital cameras snap billions of images.People are overwhelmed to manage relevant pictures among unwanted pictures.This Search Engine is developed to search user's choice of pictures among millions of images in the database, which he/she had snapped in past.

What it does

It runs deep learning algorithm LSTM offline to generate a relevant caption of the training image set.Then it creates index based out of those caption keywords for corresponding images stored in directories.Users enter certain keywords to search pictures, it looks up into index and ranks based on highest matched frequency for that image and return the results.

How I built it

I created PHP web pages.I created a database schema in Linux Ubuntu.I downloaded Flickr8k_Dataset from internet to train CNN model to generate caption keywords

Challenges I ran into

The time to train is very long.We have to wait to know if any bugs are in the program.A heuristics is implemented for ranking

Accomplishments that I'm proud of

Many mobile companies could adapt my algorithm to organize images in consumer's smartphones for their interests.

What I learned

Convolution Neural Network, Indexing, Ranking

What's next for Image Search Engine based on it's content

Human face was not mapped to particular person names.Hence, my next step is to recognize the human face and attach correct names to them as Facebook does.

Built With

• lstm
• php
• python
• scikit-learn
• unix