Name: Pavithra S G
Roll No: 247MA006
Project Progress Report on
Image Enhancement Using Super Resolution Convolutional Neural Network
Week 1: Literature Review
First week focused on studying the key concepts behind super-resolution techniques and how the Super
Resolution Convolutional Neural Network (SRCNN) works. Reviewed different methods used to enhance
low-resolution images, including traditional interpolation techniques like bicubic interpolation and learned
that SRCNN is a deep learning model specifically designed for image super-resolution. It takes a low-
resolution image and predicts a high-resolution version by learning from data.
The network has three main layers: the patch extraction layer, the non-linear mapping layer and the
reconstruction layer. Each layer has a unique function in improving image quality. For example, in SRCNN,
the patch extraction layer captures small image details, the non-linear mapping layer processes these details,
and the reconstruction layer creates the final high-resolution output. This effective architecture makes
SRCNN a powerful tool for enhancing images.
Week 2: Importing Packages and Setup
This week focused on setting up the necessary tools and libraries. Proper configuration of these packages
is essential for the smooth development and experimentation of the Super Resolution Convolutional
Neural Network (SRCNN).
1. Importing Libraries:
Importing essential Python libraries which will support various aspects including deep learning and
image processing. Here is a brief overview:
• sys: Provides access to system-specific parameters and functions.
• keras: A high-level API for building and training deep learning models.
• cv2: OpenCV library for image processing.
• numpy: Used for numerical operations and handling arrays.
• matplotlib: For creating visualizations such as plots.
• skimage: For image processing tasks.
2. Importing Modules for SRCNN:
To build and train SRCNN model, imported specific components from the Keras library:
• Sequential: For creating a linear stack of layers.
• Conv2D: For applying convolution operations in the model.
• Adam: An optimizer for training the model.
• compare_ssim: For evaluating image quality.
3. Setting Up Visualizations:
Configured the matplotlib library to display figures this allows for immediate visualization of results
and plots.
Week 3: Data Collection and Preparation:
This week focused on gathering and preparing datasets for training and testing SRCNN model. The goal
was to collect low-resolution images and prepare them for model training and evaluation. To train the
SRCNN model, needed to create low-resolution versions of the original images and achieved this by
resizing the original images using OpenCV.
The resizing process involves:
• Downscaling: Reducing the size of the image.
• Upscaling: Resizing the downscaled image back to its original size.
Used bilinear interpolation for resizing, which helps in creating smooth images by considering the closest
four pixels.
For example, for an image of size 256x256 pixels:
• Downscaling: If we reduce the size by a factor of 2, the new dimensions become 128x128 pixels.
• Upscaling: Resizing the 128x128 pixel image back to 256x256 pixels restores its original size.
Code Implementation: The following code is written to prepare the degraded images.
