USPTO Patent: Learned Image Compression Method

ChangeBridge: Patent Grants - AI & Computing (G06N)

Published March 24th, 2026

Detected March 25th, 2026

Summary

The USPTO has granted patent US12587664B2 to Sisvel Technology S.R.L. for a learned image compression method utilizing an autoencoder and entropy coding. The patent details a process involving latent space extraction, quantization, entropy coding, and reconstruction, trained via gradient descent.

View original document View source feed page

What changed

The United States Patent and Trademark Office (USPTO) has granted patent US12587664B2 to Sisvel Technology S.R.L. for a novel method of learned image compression. The patent describes an autoencoder-based approach that includes extracting a latent space from an image, quantizing it, and then entropy coding the quantized representation to produce a bitstream. The decoder reconstructs the image from this bitstream, and the entire autoencoder is trained using gradient descent to minimize a rate-distortion cost function, with a differentiable soft frequency counter for the entropy encoder.

This patent grant is primarily an intellectual property matter and does not impose direct regulatory obligations on businesses. However, companies involved in image processing, AI development, or data compression technologies should be aware of this patented technology. Companies may need to conduct freedom-to-operate analyses or consider licensing arrangements if their products or services utilize similar patented techniques. The effective date of the patent is March 24, 2026.

Source document (simplified)

← USPTO Patent Grants

Method for learned image compression and related autoencoder

Grant US12587664B2 Kind: B2 Mar 24, 2026

Assignee

Sisvel Technology S.R.L.

Inventors

Alberto Presta, Attilio Fiandrotti, Enzo Tartaglione, Marco Grangetto

Abstract

A method for learned image compression implemented in an autoencoder includes: a) extracting from an image a latent space by the learnable encoder; b) quantizing the latent space by a quantizer to obtain a quantized latent space; c) entropy coding the quantized latent space by an entropy encoder to obtain a bitstream, wherein an entropy model used to encode the latent space is represented by a probability distribution; d) entropy decoding the bitstream by an entropy decoder to obtain an entropy decoded bitstream; e) feeding the entropy decoded bitstream to the decoder; f) recover a reconstructed image by the decoder; g) training the autoencoder via standard gradient descent of the backpropagated error gradient by finding learnable parameters of the learnable encoder and of the decoder that minimize a rate distortion cost function, wherein the entropy encoder is based on a differentiable formulation of a soft frequency counter.