Image Colorizer

Overview

This project is an image colorizer web application that uses a deep learning model to colorize grayscale images. The application is built using Flask, a micro web framework in Python, and utilizes a Caffe model for the colorization process. The goal is to provide users with an easy-to-use interface for uploading grayscale images, processing them through a trained model, and downloading the colorized results.

Project Structure

• app.py: Main Flask application file handling routes and logic.
• templates/disp.html: HTML template for displaying colorized images and providing download functionality.
• static/: Directory containing static assets such as CSS, JavaScript, and images.
• uploads/: Directory where uploaded grayscale images are stored.
• downloads/: Directory where colorized images are saved after processing.
• model/: Directory containing the Caffe model files.

Caffe Model

This repository doesn't contain colorization_release_v2.caffemodel file. Download it from link to your 'model' directory.

Overview

Caffe is a deep learning framework developed by the Berkeley Vision and Learning Center (BVLC). It is known for its speed and modularity, making it suitable for a range of image processing tasks. The model used in this project is designed to colorize grayscale images.

Model Components

1. Prototxt File (colorization_deploy_v2.prototxt):

   • Defines the architecture of the neural network.
   • Specifies the layers, their connections, and other model parameters.

2. Caffemodel File (colorization_release_v2.caffemodel):

   • Contains the pre-trained weights for the neural network.
   • The weights are obtained by training the model on a large dataset of color images.

3. Numpy File (pts_in_hull.npy):

   • Contains the cluster centers used by the model for colorization.
   • These cluster centers help in mapping grayscale values to color values.

Image Colorization Process

1. Image Preprocessing:

   • Convert the input grayscale image to the LAB color space.
   • Extract the L channel (luminance) and resize the image for model input.

2. Model Inference:

   • Use the Caffe model to predict the AB channels (color) from the L channel.
   • Resize the predicted AB channels to match the dimensions of the original image.

3. Image Postprocessing:

   • Combine the L channel with the predicted AB channels.
   • Convert the LAB image back to BGR color space.
   • Save the colorized image to the specified directory.

Website Functionality

Pages

1. Index Page (index.html):

   • The landing page of the website where users can upload grayscale images.
   • Provides a form for file upload.

2. Display Page (disp.html):

   • Shows the original grayscale image and the colorized result side by side.
   • Includes a button for downloading the colorized image.

Features

• Image Upload:

  • Users can upload grayscale images through a form on the index page.
  • The uploaded image is saved in the uploads/ directory.

• Image Processing:

  • The uploaded image is processed using the Caffe model to produce a colorized version.
  • The colorized image is saved in the downloads/ directory.

• Image Download:

  • Users can download the colorized image by clicking the download button.
  • After the download, users are redirected to the index page.

Setup and Installation

Requirements

• Python 3.x
• Flask
• OpenCV
• Numpy
• Caffe

Installation Steps

1. Clone the Repository:

   git clone https://github.com/984-ISHU/Image-Colorization.git
   cd Image-Colorization