Learned Image Compression Techniques

in

Before anything else, what is image compression? It’s basically taking an image with all its juicy details and squeezing it into a smaller file size without losing too much quality. This can be useful for saving storage space or reducing bandwidth usage when transmitting images over the internet.

Now, learned image compression techniques. These are methods that use machine learning algorithms to compress images in a more efficient and effective way than traditional compression methods like JPEG or PNG. The idea is to train a neural network on a dataset of compressed images and then have it learn how to compress new images with the same level of quality but at a lower file size.

One popular learned image compression technique is called “Photographic Compression” by Google Research. This method uses a deep learning model that can predict which parts of an image are important and should be preserved, while also compressing less significant areas to save space. The result is a high-quality compressed image with up to 90% reduction in file size!

Another technique called “Learned Image Compression” by Facebook AI Research (FAIR) uses a similar approach but adds some extra bells and whistles like progressive compression, which allows for faster transmission of images over the internet. This method can achieve up to 95% reduction in file size while maintaining high image quality!

A recent paper by researchers at MIT called “Learned Image Compression with Arbitrary Bitrates” takes things a step further by allowing for customizable bitrate settings. This means you can choose how much compression to apply based on your specific needs and preferences. The result is an even more efficient use of storage space or bandwidth!

And the best part? You don’t need a degree in computer science to understand them! Just remember this: if you want high-quality compressed images with minimal file size, look for methods using machine learning algorithms like Photographic Compression or Learned Image Compression.

As for script examples, here’s some code snippets from the “Photographic Compression” paper that demonstrate how to use their method:

# Import necessary libraries
import numpy as np # Import numpy library for array manipulation
from PIL import Image # Import PIL library for image processing
import torch # Import torch library for machine learning
from torchvision.models.photometric_compression import PhotometricCompression # Import PhotometricCompression model from torchvision

# Load image and convert it into a tensor
img = Image.open('input_image.jpg') # Open input image using PIL
tensor = transforms.ToTensor()(img) # Convert image to tensor using transforms.ToTensor() function

# Initialize the compression model
model = PhotometricCompression().eval() # Initialize PhotometricCompression model and set it to evaluation mode

# Compress the input image using the model
compressed, bitstream = model(tensor) # Use the model to compress the input image and store the compressed image and bitstream

# Save compressed image and bitstream to disk
np.save('output_image', compressed.detach().cpu()) # Save the compressed image to disk using numpy's save function
with open('bitstream', 'wb') as f: # Open a file named 'bitstream' in write binary mode
    f.write(bitstream) # Write the bitstream to the file

And that’s it, You now have a high-quality compressed image with up to 90% reduction in file size using learned image compression techniques.

SICORPS