Image Conversion Unit for Code Reduction and Accuracy

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing image processing technologies face challenges in reducing code amount while maintaining image processing accuracy and human visual inspection accuracy, with schemes like pre-filtering degrading accuracy when smoothing entire images and importance maps not clearly indicating important areas for smoothing.

Innovation Solution

An image processing device and method that performs image conversion to reduce data size while maintaining feature quantities and processing accuracy, using a learning unit to optimize image conversion based on importance maps and visual accuracy maintenance, with image correction to improve processing accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of substance

If pre-filtering is applied to smooth the entire image, then code amount is reduced, but image processing accuracy is degraded

Engineering Contradiction:

Improvecode amountVSAvoidimage processing accuracy

Core Design Contradiction:

Loss of substanceVSMeasurement precision

Solution Approach 1:

The patent applies different smoothing strengths to different regions of the image based on an importance map. Regions with low importance (e.g., background areas) are smoothed more aggressively to reduce code amount, while regions with high importance (e.g., objects of interest) are smoothed less to preserve image processing accuracy. This local differentiation resolves the contradiction by optimizing the trade-off between code reduction and accuracy preservation in a spatially selective manner.

Inventive Principle:

Principle #3Local quality

2Loss of substance

If importance maps are used to guide smoothing, then code amount is reduced, but the importance maps do not clearly indicate important areas for smoothing

Engineering Contradiction:

Improvecode amountVSAvoidimportance area identification

Core Design Contradiction:

Loss of substanceVSDifficulty of detecting and measuring

Solution Approach 1:

The patent employs a feedback mechanism where the importance map is generated based on the output of a neural network that processes the image. The importance map is then used to guide the smoothing process, and the results are fed back to refine the importance map generation. This iterative feedback loop improves the clarity and accuracy of importance area identification, enabling more effective code reduction while preserving critical image features.

Inventive Principle:

Principle #23Feedback

3Manufacturing precision

If general image encoding schemes are used, then image quality is maintained, but image processing accuracy is degraded

Engineering Contradiction:

Improveimage qualityVSAvoidimage processing accuracy

Core Design Contradiction:

Manufacturing precisionVSMeasurement precision

Solution Approach 1:

The patent changes the encoding parameters by introducing an importance map-based smoothing process before encoding. Instead of using standard encoding parameters alone, the system modifies the image data according to importance regions, then applies encoding. This parameter change approach allows the system to prioritize preservation of features important for machine processing while still maintaining acceptable visual quality, thereby resolving the contradiction between image quality and processing accuracy.

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS11477460B2Image processing apparatus, learning appratus, image processing method, learning method and program

Publication Date: 2022.10.18 NIPPON TELEGRAPH & TELEPHONE CORP

AI summary

An image processing device includes an image processing unit configured to execute image processing on an image based on an input image and output a result of the image processing, the input image is a post-conversion image obtained by performing image conversion on an original image, and the conversion includes image conversion for further decreasing a data size of the original image while maintaining a feature quantity used in the image processing and processing accuracy of the image processing.