Accessible Images

• 

  Company/Team Logo

Inspiration

My cousin is legally blind and we wanted to create a tool that could help him and others who are visually impaired be able to understand text that they may not be able to read themselves.

What it does

The user takes a picture of something that they want to read the text of, uploads the picture to our website, then the AI model transcribes the image and outputs the text that it read

How we built it

We built our Convolutional Neural Network model using a TensorFlow Keras that was trained on data that we created ourselves.

Challenges we ran into

The complexity of this problem is significant so the accuracy of our model is not always accurate at identifying all of the letters in an image and also sometimes misclassifies letters.

Accomplishments that we are proud of

We feel that our project is at a state where it can realistically be used to help people read small portions of text with a high accuracy. We were able to get our model to correctly predict text from our testing dataset with a 85% accuracy, which is high for such a complex task.

What we learned

We learned a lot from this project, including how to handle and process large dataset (specifically images), train and refine a CNN ML model, and build a website that is able to apply our model to input data.

What's next for Accessible Images

There are many additions that we would have liked to add to our tool such as: Automatic text to speech after image transcription Translation of text to different languages to assist non-English readers Text transcription for other file times such as pdfs and videos

Built With

• cnn
• css
• flask
• html5
• keras
• python
• tensorflow