17987844. NEURAL IMAGE COMPRESSION WITH CONTROLLABLE SPATIAL BIT ALLOCATION simplified abstract (QUALCOMM Incorporated)

NEURAL IMAGE COMPRESSION WITH CONTROLLABLE SPATIAL BIT ALLOCATION

Organization Name

QUALCOMM Incorporated

Inventor(s)

Yang Yang of San Diego CA (US)

Hoang Cong Minh Le of La Jolla CA (US)

Yinhao Zhu of La Jolla CA (US)

Reza Pourreza of San Diego CA (US)

Amir Said of San Diego CA (US)

Yizhe Zhang of San Diego CA (US)

Taco Sebastiaan Cohen of Amsterdam (NL)

NEURAL IMAGE COMPRESSION WITH CONTROLLABLE SPATIAL BIT ALLOCATION - A simplified explanation of the abstract

This abstract first appeared for US patent application 17987844 titled 'NEURAL IMAGE COMPRESSION WITH CONTROLLABLE SPATIAL BIT ALLOCATION

Simplified Explanation

The abstract describes a method for image compression using an artificial neural network (ANN) that utilizes a spatial segmentation map to identify regions of interest. The image is compressed based on a controllable spatial bit allocation determined by a learned quantization bin size.

• The method involves receiving an image and a spatial segmentation map.
• The spatial segmentation map indicates regions of interest in the image.
• An encoder in the artificial neural network compresses the image.
• The compression is performed according to a controllable spatial bit allocation.
• The controllable spatial bit allocation is determined based on a learned quantization bin size.

Potential Applications

• Image compression for efficient storage and transmission.
• Video compression for streaming and broadcasting.
• Medical imaging for reducing data size without significant loss of information.
• Remote sensing for compressing satellite or aerial imagery.

Problems Solved

• Efficient compression of images while preserving important regions of interest.
• Reducing the size of image data for storage and transmission.
• Controlling the allocation of bits to different spatial regions based on their importance.

Benefits

• Improved storage efficiency by reducing the size of image files.
• Faster transmission of images over networks with limited bandwidth.
• Better preservation of important details in images.
• Flexibility in controlling the allocation of bits to different regions of an image.

Original Abstract Submitted

A processor-implemented method for image compression using an artificial neural network (ANN) includes receiving, at an encoder of the ANN, an image and a spatial segmentation map corresponding to the image. The spatial segmentation map indicates one or more regions of interest. The encoder compresses the image according to a controllable spatial bit allocation. The controllable spatial bit allocation is based on a learned quantization bin size.