Probabilistic ROI Image Compression for Preserving Object Details

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional image compression techniques for training automotive perception models fail to adequately preserve important details of objects of interest, leading to inaccuracies in model training due to hard boundaries and block artifacts.

Innovation Solution

Segment images into probabilistic regions with varying confidence levels of being within a region of interest and apply compression ratios inversely related to these confidence levels, with spatial smoothing across boundaries to reduce artifacts.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Quantity of substance

If conventional image compression techniques are used to reduce storage space, then storage requirements are reduced, but important details of objects of interest are lost

Engineering Contradiction:

Improvestorage spaceVSAvoidobject details

Core Design Contradiction:

Quantity of substanceVSLoss of information

Solution Approach 1:

The image is segmented into multiple regions based on probability maps generated by neural networks. Each region is assigned a compression ratio based on its likelihood of containing objects of interest, allowing differential compression that preserves important details while reducing storage space overall

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

Different regions of the image are compressed at different quality levels. Regions with high probability of containing objects of interest use lower compression ratios to preserve details, while regions with low probability use higher compression ratios to save storage space

Inventive Principle:

Principle #3Local quality

2Quantity of substance

If uniform compression is applied to entire images, then storage space is reduced, but hard boundaries and block artifacts are introduced

Engineering Contradiction:

Improvestorage spaceVSAvoidimage quality

Core Design Contradiction:

Quantity of substanceVSManufacturing precision

Solution Approach 1:

The image is divided into multiple regions based on probability maps. Each region is then compressed independently at an optimized compression ratio, avoiding the hard boundaries and block artifacts that occur with uniform compression while maintaining storage efficiency

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

Different compression ratios are applied to different regions based on their content characteristics. This local differentiation allows each region to be compressed optimally without introducing artifacts, as the compression parameters adapt to the specific needs of each region

Inventive Principle:

Principle #3Local quality

Data Source

PatentUS20250378585A1Content-based image compression via probabilistic region of interest segmentation

Publication Date: 2025.12.11 QUALCOMM INC

AI summary

A computing system may segment an image frame into three or more probabilistic regions, wherein each of the three or more probabilistic regions is associated with a corresponding confidence level of being within a region of interest in the image frame out of a plurality of confidence levels of being within the region of interest. A computing system may determine, for each of the three or more probabilistic regions, a corresponding compression ratio based on the corresponding confidence level of being within the region of interest. A computing system may compress each of the three or more probabilistic regions according to the corresponding compression ratio.