Image Hash Code Retaining Spatial Information for Local Matching

Solution Overview

Problem

Current image hashing technologies struggle to effectively handle local matching of images, as they fail to capture spatial information, leading to difficulties in recognizing similarities between images with partially offset similar regions.

Innovation Solution

The method involves obtaining image features through a trained neural network, segmenting them into local regions, determining local feature vectors, mapping these vectors into sub-hash codes, and combining them to form a hash code that retains spatial information, allowing for accurate local image matching by arranging sub-hash codes according to their spatial arrangement.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of information

If traditional image hashing methods generate a binary vector where each bit condenses information of the entire image, then the hash code achieves compact representation and fast computation, but it loses spatial information and cannot effectively handle local matching of images

Engineering Contradiction:

Improvespatial informationVSAvoidhash code structure

Core Design Contradiction:

Loss of informationVSDevice complexity

Solution Approach 1:

The image feature is divided into multiple local regions, and each local region generates a separate sub-hash code. This segmentation preserves spatial information by maintaining the correspondence between local regions and their positions in the original image, while avoiding the information loss of global pooling.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent transforms the traditional one-dimensional binary hash vector into a two-dimensional matrix structure where rows represent different local regions and columns represent hash bits. This dimensional change allows spatial information to be preserved in the matrix structure while maintaining compact representation.

Inventive Principle:

Principle #17Another dimension (Dimensionality change)

2Measurement precision

If the image feature is divided into multiple local regions and each region generates a sub-hash code, then spatial information is preserved for local matching, but the hash code structure becomes more complex

Engineering Contradiction:

Improvelocal matching accuracyVSAvoidhash code structure

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The image feature is divided into multiple local regions, and each local region generates a separate sub-hash code. This segmentation preserves spatial information by maintaining the correspondence between local regions and their positions in the original image, while avoiding the information loss of global pooling.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

Multiple sub-hash codes from different local regions are combined into a single hash code matrix through concatenation along the row dimension. This merging process integrates local features while preserving their spatial arrangement, achieving both local matching capability and unified representation.

Inventive Principle:

Principle #5Merging (Combining)

3Productivity

If each bit in the hash code represents global image information, then the hash code achieves simplicity and fast computation, but it fails to capture local similarities between images with partially offset regions

Engineering Contradiction:

Improvecomputation speedVSAvoidlocal similarity detection

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The image feature is divided into multiple local regions, and each local region generates a separate sub-hash code. This segmentation preserves spatial information by maintaining the correspondence between local regions and their positions in the original image, while avoiding the information loss of global pooling.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

Instead of requiring complete global similarity, the method uses partial matching by comparing local regions independently. This allows the system to identify images with partial overlaps or locally similar regions, achieving more flexible and accurate image retrieval.

Inventive Principle:

Principle #16Partial or excessive action

Data Source

PatentUS11714921B2Image processing method with ash code on local feature vectors, image processing device and storage medium

Publication Date: 2023.08.01 BOE TECHNOLOGY GROUP CO LTD

AI summary

Provided are an image processing method, an image matching method, a device, and a storage medium. The image processing method includes: obtaining an image feature of an input image; determining a plurality of local image features of the image feature; determining a plurality of local feature vectors corresponding to the plurality of local image features respectively; determining a hash code of the input image based on the plurality of local feature vectors.