Kernel-Based Signal Coding for Low-Entropy Machine Vision Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current coding techniques for image, video, and audio signals are primarily designed for human consumption, discarding higher frequencies, which is inadequate for machine processing that requires different signal transformations.
Innovation Solution
An encoder system with a transformer module that applies entropy-minimizing kernel-based transformations to input signals, reducing their entropy and making them suitable for machine processing, coupled with a compressor module that generates a compressed bitstream for transmission.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional coding techniques are used that are designed for human consumption, then the signal is optimized for human visual system processing, but the signal is not suitable for machine processing which requires different frequency information
Solution Approach 1:
The patent changes the transformation parameters from human-optimized DCT bases to machine-optimized kernel bases that preserve high frequency information. The transform kernels are specifically designed to maintain information relevant to machine tasks such as object detection and classification, rather than optimizing for human visual perception.
Solution Approach 2:
The patent introduces dynamic adaptability by allowing the transform kernels to be selected or designed based on the specific machine task requirements. Different kernels can be used for different machine processing tasks, making the coding system dynamically adaptable rather than static and human-centric.
2Productivity
If signal transformation is applied to reduce entropy for compression, then compression efficiency is improved, but the transformation complexity increases
Solution Approach 1:
The patent extracts only the essential transformation operations needed for machine processing and compression. Rather than applying complex multi-stage transformations, it focuses on the key kernel-based transformation that simultaneously achieves both machine suitability and entropy reduction for compression.
3Loss of energy
If high frequency information is discarded to optimize for human visual system, then bandwidth is reduced, but machine processing capability is degraded
Solution Approach 1:
The patent applies local quality optimization by preserving high frequency information in regions and at resolutions that are specifically useful for machine processing tasks. Rather than uniformly discarding high frequencies, the transformation maintains local details that are critical for machine analysis while still achieving compression.
Data Source
AI summary
A system for coding and decoding of signals, such as pictures and video, intended for machine use is disclosed. An encoder implements a method for kernel-based transformation of the signal samples that reduces and preferably minimizes entropy of the input signal yet is equivalent to the input signal for the intended machine operation. As a result, the transformed signal sample can be compressed using fewer bits than the original. The encoded signal is received by a complimentary decoder and decoded for machine use.