Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Concept Videos

Force Classification01:22

Force Classification

Forces play a crucial role in the study of physics and engineering. They are essential in describing the motion, behavior, and equilibrium of objects in the physical world. Forces can be classified based on their origin, type, and direction of action.
Contact and non-contact forces are two of the most widely used categories of forces. As the name suggests, contact forces require physical contact between two objects to act upon each other. Examples of contact forces include frictional,...
Extraction: Advanced Methods00:56

Extraction: Advanced Methods

Metal ions can be separated from one another by complexation with organic ligands–the chelating agent– to form uncharged chelates. Here, the chelating agent must contain hydrophobic groups and behave as a weak acid, losing a proton to bind with the metal. Since most organic ligands used in this process are insoluble or undergo oxidation in the aqueous phase, the chelating agent is initially added to the organic phase and extracted into the aqueous phase. The metal-ligand complex is formed in...
Deconvolution01:20

Deconvolution

Deconvolution, also known as inverse filtering, is the process of extracting the impulse response from known input and output signals. This technique is vital in scenarios where the system's characteristics are unknown, and they must be inferred from the observable signals.
Deconvolution involves several mathematical techniques to derive the impulse response. One common approach is polynomial division. In this method, the input and output sequences are treated as coefficients of...

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

Predictors of clomiphene citrate response in the treatment of men with testosterone deficiency.

The journal of sexual medicine·2026
Same author

Patterns of mechanical failure in the Coloplast Titan Touch inflatable penile prosthesis: a 12-year single-surgeon experience.

The journal of sexual medicine·2026
Same author

Novel transformer-based model for NID in fog computing environment.

Scientific reports·2026
Same author

The expression profiling of serum circPHLPP2 and LncRNA ILF3 in colorectal cancer patients.

Scientific reports·2026
Same author

Dysregulation of Niacin-Derived NAD<sup>+</sup> Salvage Pathway Markers (CD38, NAMPT, SIRT1) Across Albuminuria Stages in Type 2 Diabetes.

Medicina (Kaunas, Lithuania)·2025
Same author

Rapid reagent free COVID19 detection using MEMS based FTIR spectroscopy and machine learning in NIR and MIR regions.

Scientific reports·2025

Related Experiment Videos

A hybrid ConvNeXt-BiLSTM framework for robust scene text recognition.

Alshefaa Khattab1, Marwa Elpeltagy2, Farida Youness1

  • 1Department of Systems and Computers, Al-Azhar University, Cairo, Egypt.

Scientific Reports
|May 13, 2026
PubMed
Summary
This summary is machine-generated.

This study introduces a ConvNeXt and BiLSTM deep learning framework for Scene Text Recognition (STR), achieving 94.71% accuracy on benchmarks. The model excels in complex environments, outperforming previous methods and demonstrating real-time deployment suitability.

Keywords:
Convolutional network nextDeep learningFocal lossLabel smoothingScene text recognition

Related Experiment Videos

Area of Science:

  • Computer Vision
  • Deep Learning
  • Natural Language Processing

Background:

  • Scene Text Recognition (STR) is crucial for applications like autonomous navigation and document digitization.
  • Existing STR models struggle with generalization due to reliance on synthetic data and scarcity of real-world annotated data.
  • Complex environments with irregular text, multilingual content, and varied backgrounds pose significant challenges for current STR systems.

Purpose of the Study:

  • To develop a robust deep learning framework for Scene Text Recognition (STR) that overcomes limitations of existing models.
  • To enhance the generalization capability of STR models in complex real-world scenarios.
  • To improve training stability and address class imbalance and overconfidence in STR.

Main Methods:

  • A ConvNeXt-based deep learning framework integrating Convolutional Network Next (ConvNeXt) for feature extraction and Bidirectional Long Short-Term Memory (BiLSTM) for sequence modeling.
  • Implementation of label smoothing and focal loss to improve training stability and mitigate class imbalance and overconfidence.
  • A two-stage training strategy involving pre-training on synthetic datasets (MJSynth, SynthText) followed by fine-tuning on diverse real-world datasets (IC13, IC15, RCTW, ArT, LSVT, MLT19, ReCTS, COCO-Text, Uber-Text, TextOCR, OpenVINO, Union14M-L subset).

Main Results:

  • The proposed model achieved an average accuracy of 94.71% across six standard STR benchmarks (IIIT5k, SVT, IC13, IC15, SVTP, CUTE80).
  • Performance significantly improved compared to models trained solely on synthetic data (89.1% accuracy).
  • The framework demonstrated superior performance over state-of-the-art methods under comparable data conditions, especially in challenging scenarios.

Conclusions:

  • The ConvNeXt and BiLSTM framework, combined with advanced loss functions and heterogeneous datasets, substantially enhances STR performance.
  • The model shows practical suitability for real-time deployment with 20.3M parameters, 1.9 GFLOPs, and 2.638 ms inference latency per image.
  • The proposed approach effectively addresses the generalization limitations of traditional STR models in complex and diverse environments.