Compression Neural Network Padding and Trimming for Efficient Image Coding

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional data compression networks face inefficiencies due to increased padding areas as the ratio of downscaling/upscaling increases, deteriorating compression efficiency.

Innovation Solution

Implement multiple padding and downscaling operations in an encoding apparatus, followed by corresponding upscaling and trimming operations in a decoding apparatus, to manage image size adjustments efficiently.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of substance

If multiple downscaling operations are applied to compress image data, then compression ratio is improved, but padding area increases deteriorating compression efficiency

Engineering Contradiction:
Improvecompression efficiencyVSAvoidpadding area
Core Design Contradiction:
Loss of substanceVSArea of stationary object

Solution Approach 1:

The patent segments the single downscaling operation into multiple sequential downscaling operations (e.g., first downscaling by factor of 2, then second downscaling by factor of 2 again to achieve total 4x compression). Correspondingly, padding is applied at each intermediate stage rather than once at the beginning. This segmentation allows the system to maintain smaller padding areas at each stage compared to applying all padding upfront, thereby improving overall compression efficiency while achieving the same total compression ratio.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies padding operations preliminarily at each downscaling stage rather than all at once before compression. By performing partial padding operations at intermediate points during the multi-stage downscaling process, the system prepares just enough data for each subsequent operation without unnecessarily expanding the total padding area, thus optimizing compression efficiency.

Inventive Principle:
Principle #10Preliminary action

2Productivity

If downscaling ratio is increased to improve compression, then compression efficiency deteriorates due to increased padding area

Engineering Contradiction:
Improvecompression efficiencyVSAvoidoperation complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The patent divides a single high-ratio downscaling operation into multiple lower-ratio sequential operations (e.g., splitting 4x downscaling into two 2x operations with intermediate padding). This segmentation reduces the padding area required at each stage while achieving the same overall compression, thereby improving compression efficiency without excessively increasing system complexity.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent implements periodic padding operations at regular intervals during the downscaling process. Instead of applying all padding at the beginning or using irregular padding schedules, the system applies padding periodically at each downscaling stage, creating a rhythmic pattern of compression and padding that optimizes efficiency while maintaining manageable operational complexity.

Inventive Principle:
Principle #19Periodic action

Data Source

PatentUS12417557B2Method, apparatus, and storage medium using padding/trimming in compression neural network
Publication Date: 2025.09.16 ELECTRONICS & TELECOMM RES INST
  • US12417557B2 patent drawing
  • US12417557B2 patent drawing
  • US12417557B2 patent drawing

AI summary

An encoding apparatus extracts features of an image by applying multiple padding operations and multiple downscaling operations to an image represented by data and transmits feature information indicating the features to a decoding apparatus. The multiple padding operations and the multiple downscaling operations are applied to the image in an order in which one padding operation is applied and thereafter one downscaling operation corresponding to the padding operation is applied. A decoding method receives feature information from an encoding apparatus, and generates a reconstructed image by applying multiple upscaling operations and multiple trimming operations to an image represented by the feature information. The multiple upscaling operations and the multiple trimming operations are applied to the image in an order in which one upscaling operation is applied and thereafter one trimming operation corresponding to the upscaling operation is applied.