Deep Learning In-Loop Filter for Video Encoding

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current video encoding techniques face challenges in achieving high encoding efficiency and image enhancement due to increasing image size, resolution, and frame rate, which result in higher data amounts, necessitating improved compression methods.

Innovation Solution

A video encoding method and decoding method that utilize a deep learning-based in-loop filter to detect a reference region from a current frame and a reference frame, combining the detected region with the current frame to enhance image quality and improve encoding efficiency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Manufacturing precision

If deep learning-based detection model is applied to detect reference regions, then image quality is enhanced and encoding efficiency is improved, but device complexity and computational resources increase

Engineering Contradiction:

Improveimage qualityVSAvoiddevice complexity

Core Design Contradiction:

Manufacturing precisionVSDevice complexity

Solution Approach 1:

The patent divides the reference frame into multiple regions by detecting reference regions using a deep learning-based detection model. This segmentation allows selective combination of only relevant reference regions with the current frame, improving image quality while managing computational complexity through targeted processing rather than full-frame processing.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent applies different processing quality to different regions by detecting reference regions and combining them selectively with the current frame based on a detection map. This local quality approach enhances important regions while maintaining acceptable quality elsewhere, optimizing the balance between image quality enhancement and computational resources.

Inventive Principle:

Principle #3Local quality

2Manufacturing precision

If image size, resolution, and frame rate are increased, then video quality is improved, but data amount increases requiring more compression

Engineering Contradiction:

Improvevideo qualityVSAvoiddata amount

Core Design Contradiction:

Manufacturing precisionVSQuantity of substance

Solution Approach 1:

The patent extracts and utilizes useful information from reference frames by detecting reference regions that contain valuable data. By identifying and combining only the relevant reference regions with the current frame, the system recovers important visual information that would otherwise be lost during compression, thereby improving video quality without proportionally increasing data amount.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent creates a composite image by combining the current frame with detected reference regions from reference frames. This composite approach merges information from multiple sources (current frame data and historical reference data) to produce enhanced video output with better quality than either source alone, effectively increasing video quality without linearly increasing data transmission requirements.

Inventive Principle:

Principle #40Composite materials

Data Source

PatentUS20230269399A1Video encoding and decoding using deep learning based in-loop filter

Publication Date: 2023.08.24 HYUNDAI MOTOR CO LTD

AI summary

A video encoding method and a video decoding method is provided for generating improved picture quality for a current frame and improving encoding efficiency. The video encoding method and the video decoding method further include an in-loop filter that detects a reference region from a current frame and a reference frame using a deep learning-based detection model and then combines the detected reference region with the current frame.