Pre-Analysis Image Compression for Machine Vision Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image/video compression techniques are inadequate for machine vision tasks as they focus on human perception rather than machine understanding, leading to inefficient data transmission and storage.
Innovation Solution
Perform pre-analysis based image data processing that includes instance segmentation to generate segment masks, merge these masks, extract relevant regions, and compress the image data accordingly.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If conventional image/video compression techniques are used, then transmission and storage efficiency is improved, but machine vision task performance deteriorates because the techniques focus on human perception rather than machine understanding
Solution Approach 1:
The patent applies segmentation by dividing the image into multiple regions based on object instances and their semantic importance. The encoder performs instance segmentation to identify distinct objects and segments the image into regions of interest (ROIs) and non-ROIs. This allows differential compression where ROIs maintain higher quality for machine vision tasks while non-ROIs are compressed more aggressively, resolving the contradiction between compression efficiency and machine understanding quality.
Solution Approach 2:
The patent implements local quality by applying different compression quality levels to different regions of the image. Regions containing important objects or features relevant to machine vision tasks are encoded with higher quality, while less important regions use lower quality compression. This region-dependent quality approach ensures that machine vision performance is maintained in critical areas while achieving overall compression efficiency.
2Quantity of substance
If image data is compressed to reduce data volume, then transmission and storage efficiency is improved, but the quality of image data for machine vision tasks deteriorates
Solution Approach 1:
The patent segments the image data into multiple regions with different compression requirements. By identifying regions of interest through instance segmentation and semantic analysis, the system applies appropriate compression levels to each segment. This allows significant data volume reduction in non-critical regions while preserving quality in regions essential for machine vision tasks, thus resolving the contradiction between data volume and quality.
Solution Approach 2:
The patent implements local quality control by encoding different regions with different quality parameters. Regions containing objects important for machine vision tasks are encoded with higher quality settings, while background or less important regions use lower quality settings. This selective quality approach maintains manufacturing precision (image quality) where needed while reducing overall data volume.
3Manufacturing precision
If conventional compression techniques focusing on human perception are applied, then visual quality for human viewers is improved, but efficiency for machine vision tasks deteriorates
Solution Approach 1:
The patent inverts the traditional compression approach by optimizing for machine vision tasks rather than human perception. Instead of using human visual system characteristics as the basis for compression, the system uses machine vision task requirements to guide compression decisions. This inversion allows the system to prioritize preservation of features and objects that are critical for machine analysis, thereby improving machine vision task efficiency while maintaining acceptable visual quality.
Solution Approach 2:
The patent changes the optimization parameters from human perception metrics to machine vision performance metrics. The encoding process adjusts quality parameters, region importance weights, and compression settings based on the specific requirements of machine vision tasks such as object detection, tracking, or recognition. This parameter change enables the system to optimize productivity for machine vision tasks while maintaining sufficient visual quality.
Data Source
AI summary
The present disclosure provides pre-analysis based methods for adaptively compressing image data consumed by machine vision tasks. An exemplary method includes: receiving a video sequence; encoding one or more input pictures associated with the video sequence; and generating a bitstream, wherein the encoding includes: performing instance segmentation of an input picture, to generate one or more segment masks; combining the one or more segment masks to generate a merged mask; extracting, from the input picture, a region comprising the merged mask; and compressing image data representing the extracted region.