You are on page 1of 1

Thesis Disposition:

Title: Interpretation and interpolation of pixel values - Improvement of image


scaling methods.
Image scaling refers to changing the size of an image while maintaining its aspect
ratio. This can be done using different methods, including interpolation or
resampling.
Scaling an image can impact image quality both positively and negatively. When
scaling up an image, interpolation is utilized to fill the missing pixels, which could
result in a loss of sharpness and details. However, scaling down an image uses
resampling to eliminate pixels, which might lead to the loss of information and
details.
Despite its potential negative effects, image scaling can have practical
applications. For instance, reducing the size of images can optimize web content
or decrease file size, while increasing the size of images can be helpful for
printing or display purposes.
The goal of this thesis is to utilize machine learning (ML) based image scaling
methods to produce better outcomes and evaluate the results.
Pixel Value Interpretation: Proper pixel value interpretation is essential to
preserve image details when scaling images. It helps maintain image quality by
understanding pixel values and their significance in the scaled image.
• Interpolation Methods: Interpolation fills in missing pixel values to
maintain aspect ratios. However, methods such as bicubic, bilinear, and
nearest neighbor have limitations. Proper selection of the appropriate
method is important to preserve image quality.
• ML-Based Techniques: ML-based techniques like Generative Adversarial
Networks (GANs) and Convolutional Neural Networks (CNNs) can improve
image scaling. GANs use a generator and a discriminator to create high-
resolution images, while CNNs extract features from low-resolution
images to generate high-resolution images.
This analysis highlights the significance of selecting appropriate pixel
interpolation techniques or ML-based methods to improve image scaling quality.
ML-based methods can offer superior results, and metrics such as SSIM and
PSNR, along with perceptual learning, aid in assessing and selecting optimal
scaling techniques for diverse applications.

You might also like