Doodle Recognition — 5-Class CNN An

Doodle Recognition — 5-Class CNN

An interactive computer vision project that classifies hand-drawn sketches into five categories using a Convolutional Neural Network (CNN) trained on Google's Quick, Draw! dataset.

Overview

The project explores how deep learning can recognize simple drawings from their visual patterns. A CNN learns features such as edges, curves, and shapes to distinguish between five classes:

Apple

Cookie

Dog

Eye

Face

The goal is to connect model training with an interactive drawing experience, allowing users to submit sketches and receive a predicted class.

Technical Approach

Data preparation: Process the drawing data into model-ready arrays and corresponding class labels.

Model development: Train a CNN using TensorFlow/Keras for five-class image classification.

Inference: Preprocess user drawings consistently with the training data before passing them to the trained model.

User interface: Provide a drawing canvas and display the model's predicted class.

Tech Stack

Python

TensorFlow / Keras

NumPy

Google Quick, Draw! dataset

What This Project Demonstrates

CNN-based image classification

Image preprocessing and labelled dataset preparation

Multi-class prediction

Integration of a trained model into an interactive application

Separation of model training, inference, and user-interface components

Source Code