Image Text Recognition Using Relative Position Encoding

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing technologies struggle to accurately identify a large number of character strings relevant to a target object from a target image, particularly when the target object is associated with a character string in a different small area.

Innovation Solution

The information processing device divides a target image into multiple small areas, recognizes character strings and their positions, assigns indices based on relative positional relationships, generates input data by combining word features with positional features, and processes this data through a language model to identify relevant character strings.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If the target image is divided into multiple small areas for processing, then the processing complexity is reduced and local character recognition is improved, but the ability to associate character strings across different small areas is lost

Engineering Contradiction:

Improvecharacter string recognition accuracyVSAvoidpositional relationship information

Core Design Contradiction:

Measurement precisionVSLoss of information

Solution Approach 1:

The target image is divided into multiple small areas to reduce processing complexity and improve local character recognition accuracy. Each small area can be processed independently by the recognition unit, making the overall processing more manageable and efficient.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent adds a temporal dimension by sequentially processing small areas and integrating results. The recognition unit processes each small area in sequence and the integration unit combines these results, effectively adding a time dimension to the spatial segmentation to recover the lost global contextual information.

Inventive Principle:

Principle #17Another dimension (Dimensionality change)

2Productivity

If character recognition is performed on each small area independently, then processing efficiency is improved, but the number of relevant character strings identified is reduced

Engineering Contradiction:

Improveprocessing efficiencyVSAvoidcompleteness of character string identification

Core Design Contradiction:

ProductivityVSReliability

Solution Approach 1:

The division unit performs preliminary segmentation of the target image into multiple small areas before recognition. This preliminary action allows the recognition unit to process each area efficiently while the subsequent integration step ensures that all relevant character strings are identified by combining results from all small areas.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The integration unit acts as a feedback mechanism that aggregates recognition results from all small areas. By collecting and integrating results from each independently processed small area, the system ensures that no relevant character strings are missed, thereby maintaining high reliability while preserving processing efficiency.

Inventive Principle:

Principle #23Feedback

3Device complexity

If only character strings directly associated with the target object are identified, then processing simplicity is maintained, but the completeness of information extraction is reduced

Engineering Contradiction:

Improveprocessing complexityVSAvoidnumber of character strings identified

Core Design Contradiction:

Device complexityVSQuantity of substance

Solution Approach 1:

The recognition unit is designed to be universal, processing character strings in any small area regardless of direct association with the target object. The integration unit then filters and organizes these results to identify both directly associated and indirectly relevant character strings, maintaining processing simplicity while increasing information completeness.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Solution Approach 2:

The integration unit merges results from all small areas to identify a comprehensive set of relevant character strings. By combining recognition results across all segmented areas and applying integration logic, the system identifies both directly associated and indirectly relevant character strings without significantly increasing processing complexity.

Inventive Principle:

Principle #5Merging (Combining)

Data Source

PatentUS20260057688A1Information processing apparatus, information processing method, and non-transitory computer readable medium

Publication Date: 2026.02.26 NEC CORP

AI summary

An information processing device includes a division unit, a recognition unit, an index assignment unit, an input data generation unit, and an output data acquisition unit. The division unit divides a target image including a target object and multiple character strings into multiple small areas. The recognition unit recognizes the character strings by performing character recognition processing using the target image, and recognizes positions of the character strings. The index assignment unit assigns, to each small area, an index associated with a relative positional relationship of the small areas. The input data generation unit generates input data including an input feature in which a positional feature obtained by encoding the index is added to a word feature extracted from each character string. The output data acquisition unit obtains output data obtained by inputting the input data to a language model.