Fractional Pixel Reference Generation for Video Encoding Memory

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

In video compression techniques, the precision of motion search is restricted by the resolution of the motion vector, leading to excessive memory consumption and processing delays when encoding partial regions by separate chips, especially in high-resolution or high-frame-rate video processing.

Innovation Solution

An image processing apparatus and method that generates reference pixels at fractional pixel positions further inward than a predetermined distance from the boundary of a partial region, reducing the need for external pixel values and minimizing inter-chip communication by excluding reference pixels outside this distance from the search range.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If reference pixels at fractional pixel positions are generated using interpolation filters for motion compensation, then prediction precision is improved, but memory consumption increases excessively when encoding partial regions by separate chips

Engineering Contradiction:
Improveprediction precisionVSAvoidmemory consumption
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The image is divided into multiple partial regions (tiles or slices) that can be processed independently by separate chips. By segmenting the processing domain, each chip only needs to generate and store reference pixels for its own partial region, dramatically reducing the memory burden while maintaining prediction precision within each segment

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies different processing strategies to different spatial locations. For pixels near the boundary of partial regions, the patent restricts the use of filter taps that would extend into adjacent regions, while allowing full interpolation filter usage for pixels away from boundaries. This local differentiation optimizes memory usage without sacrificing overall prediction quality

Inventive Principle:
Principle #3Local quality

2Measurement precision

If reference pixels outside the target region are stored in memory for motion search, then motion compensation precision is improved, but processing delay increases due to frequent inter-chip communication

Engineering Contradiction:
Improvemotion compensation precisionVSAvoidprocessing delay
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

By dividing the image into independent partial regions processed by separate chips, the patent eliminates the need for frequent inter-chip communication to access reference pixels. Each chip maintains only the reference pixels needed for its own partial region, making the system scalable and reducing communication overhead as the number of chips increases

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent generates reference pixels at fractional pixel positions in advance for each partial region before motion search begins. By pre-computing these reference pixels within the bounds of each partial region, the system avoids the need for real-time inter-chip communication during the motion search process, significantly reducing processing delay

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentUS10743023B2Image processing apparatus and image processing method
Publication Date: 2020.08.11 SONY GROUP CORP
  • US10743023B2 patent drawing
  • US10743023B2 patent drawing
  • US10743023B2 patent drawing

AI summary

An image processing apparatus including an inter-prediction section that generates a set of reference pixels of fractional pixel positions from a reference image for motion compensation and searches an image to be encoded for a motion vector by using the set of generated reference pixels and an encoding section that encodes the image to be encoded on a basis of a result of searching for the motion vector by the inter-prediction section. The inter-prediction section includes a reference pixel at a fractional pixel position further inward than a first predetermined distance from a boundary of a partial region corresponding to a slice or a tile, in the set of reference pixels that are used for searching the partial region for a motion vector.