AI Image Encoding with Region-Specific Compression Recovery
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing encoding methods fail to maintain image quality for regions other than the target region in AI recognition processes, particularly when a uniform compression rate is applied across all regions, leading to reduced convenience in decoding data usage.
Innovation Solution
An encoding apparatus that determines a limit compression rate for target regions in image data, encodes all regions at this rate, generates first invalidated data by excluding non-target regions, and encodes difference data at a predetermined rate, while a decoding apparatus reconstructs the image quality by adding decoded difference data to the target region.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of substance
If a uniform compression rate is applied across all regions, then the data size is reduced, but the image quality for regions other than the target region deteriorates
Solution Approach 1:
The image is divided into a target region (containing the recognition target) and a non-target region. Different compression rates are applied to each region: a first compression rate to the target region and a second compression rate (lower than the first) to the non-target region. This segmentation allows optimized compression for each area, maintaining recognition accuracy while preserving non-target region quality.
Solution Approach 2:
Different compression qualities are applied to different parts of the image based on their importance. The target region receives higher compression since it only needs to maintain recognition accuracy, while the non-target region receives lower compression to preserve visual quality for other potential uses. This local quality differentiation resolves the contradiction between overall data reduction and regional image quality.
2Loss of energy
If high compression rate is applied to reduce data size, then transmission cost is reduced, but the convenience of decoding data usage for non-AI tasks deteriorates
Solution Approach 1:
The encoded data is segmented into two parts with different compression levels: highly compressed target region data and lightly compressed non-target region data. This allows the system to transmit less overall data (reducing transmission cost) while ensuring that the non-target region maintains sufficient quality for general viewing and other applications beyond AI recognition.
Solution Approach 2:
The compression rate parameter is changed based on spatial location within the image. By adjusting the compression parameter differently for target and non-target regions, the system optimizes the balance between transmission efficiency and decoding data usability for various purposes.
3Manufacturing precision
If low compression rate is applied to maintain image quality, then image quality is preserved, but the data size increases
Solution Approach 1:
The image is segmented into target and non-target regions, allowing selective application of compression rates. The non-target region receives lower compression (preserving quality) while the target region receives higher compression (reducing data size). This segmentation strategy maintains overall image quality where needed while minimizing total data size.
Solution Approach 2:
Different quality levels are applied locally to different regions based on their functional requirements. The non-target region maintains higher quality for general viewing purposes, while the target region uses lower quality sufficient for AI recognition, optimizing the quality-to-size ratio across the entire image.
Data Source
AI summary
An encoding apparatus includes: a memory; and a processor coupled to the memory and configured to: determine, based on a result of a recognition process by AI, a target region desirable for recognizing a recognition target in image data and a limit compression rate at which the recognition target is recognizable; encode all regions of the image data at the limit compression rate, and transmits the encoded image data; generate first invalidated image data by invalidating a region other than the target region in the image data, and second invalidated image data by invalidating the region other than the target region in decoding data obtained by encoding the image data at the limit compression rate, and decoding the encoded image data; and encode difference data between the first invalidated image data and the second invalidated image data at a predetermined compression rate, and transmits the encoded difference data.