CanIGetTheNumber

Code uploaded on GitHub
Model Accuracy
WebPage Output

Inspiration

At times, we find it difficult to read the handwritten phone numbers. Most of the times you end up calling the wrong person since the number given to you in the written form is tricky to read.

What it does

This model focuses on giving best accuracy in detecting the numeric values correctly. We need to take a picture of the sheet of paper and upload it on the handy website we provide.

How we built it

We use Tensorflow and Keras to build the ML model. For outputting the prediction, we have a friendly UI.

Challenges we ran into

For improving the accuracy, different analysis was required. Converting the images into black and white, then gray scale, cropping and sharpening the images. Basically, all the image processing part was crucial in finding the best accuracy.

Accomplishments that we're proud of

We were able to give an accuracy of 0.9% for detecting the numbers from handwritten images. Connecting the ML model with flask to output the results on Web based UI was the most interesting part. We faced issues while running flask and rendering the website on real time basis results. The model not only predicts from the pre given mnist database, but also predicts well for the new data that we provide on the web UI in real time.

What we learned

Learned different algorithms like CNN, Tensorflow and Keras. We also learned about Flask which is very useful as a connection between the backend and frontend.

What's next for CanIGetTheNumber

Currently the project focuses on building good accuracy. Thus, we only have single digits being detected. Later we can have the entire 10 digit number detected and automatically stored in your mobile contacts through a database connection with Firebase/MongoDB