SIFT Descriptor Compression via SVD for Mobile Image Analysis

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image analysis systems face inefficiencies due to the high amount of data required for processing and transmission, particularly in mobile terminals, where the large number of keypoints and SIFT descriptors lead to excessive computational load and data transmission, limiting scalability and performance.

Innovation Solution

The method involves compressing the codebook matrix through factorization and truncation, followed by quantization and scaling, to reduce the data required for processing and storage, specifically using Singular Value Decomposition (SVD) to concentrate information in fewer columns and representing data with optimized quantization levels, thereby reducing the memory space occupied by the codebook.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If SIFT descriptors are used for accurate image comparison, then measurement precision is improved, but quantity of substance increases due to large data size

Engineering Contradiction:
Improveimage comparison accuracyVSAvoiddata amount
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent extracts and transmits only the essential information from SIFT descriptors - specifically the support values and their positions - while discarding redundant data. This allows maintaining image comparison accuracy while significantly reducing the quantity of data that needs to be processed and transmitted.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

Instead of transmitting the complete SIFT descriptor data and then processing it, the patent inverts the approach by first identifying and extracting only the critical support values and their positions, then transmitting this condensed information. This reversal of the traditional workflow eliminates unnecessary data transmission while preserving measurement precision.

Inventive Principle:
Principle #13The other way round (Inversion)

2Reliability

If all keypoint data is transmitted from mobile terminal to server, then reliability is improved, but loss of energy increases due to excessive data transmission

Engineering Contradiction:
Improveservice reliabilityVSAvoiddata transmission energy
Core Design Contradiction:
ReliabilityVSLoss of energy

Solution Approach 1:

The patent extracts only the essential support values and their positions from the complete keypoint data, transmitting only this condensed information from mobile terminal to server. This extraction approach maintains service reliability by preserving the critical information needed for accurate image comparison while dramatically reducing the energy consumed during data transmission.

Inventive Principle:
Principle #2Taking out (Extraction)

3Measurement precision

If complete SIFT descriptor data is processed, then measurement precision is improved, but productivity decreases due to excessive computational load

Engineering Contradiction:
Improveimage analysis accuracyVSAvoidprocessing efficiency
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent extracts and processes only the support values and their positions from the complete SIFT descriptor data, eliminating the need to process the full 128-byte descriptors. This extraction approach maintains image analysis accuracy by preserving the essential information while significantly improving processing efficiency by reducing computational load.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent applies partial action by processing only the necessary portion of the SIFT descriptor data (support values and positions) rather than the complete descriptor. This partial processing approach is sufficient to achieve accurate image comparison while dramatically improving productivity by avoiding unnecessary computational operations on redundant data.

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP2842106B1Method and system for image analysis
Publication Date: 2019.11.13 TELECOM ITALIA SPA
  • EP2842106B1 patent drawingFigure 1
  • EP2842106B1 patent drawingFigure 2A
  • EP2842106B1 patent drawingFigure 2B

AI summary

A method for processing an image is disclosed. The method comprises: • - identifying a group of keypoints in the image; • - for each keypoint of the group; • • a) calculating a corresponding descriptor array including a plurality of array elements, each array element storing values taken by a corresponding color gradient histogram of a respective sub-region of the image in the neighborhood of the keypoint; • b) generating at least one compressed descriptor array by compressing at least one portion of the descriptor array by means of vector quantization using a codebook (CBK) comprising a plurality of codewords (CW).