Hybridized Dimensionality Reduction Model for Blurred Text Detection in Natural Scene Images

Simple item page
dc.contributor.authorChinonyelum Vivian NWUFOH
dc.date.accessioned2024-10-15T11:10:09Z
dc.date.available2024-10-15T11:10:09Z
dc.date.issued2023-12
dc.description.abstractScene Text Recognition (STR) is synonymous with text recognition in the wild scene, and it is a difficult task since it necessitates getting rid of text strokes and their fuzzy borders, such as embossing, shading, and flare, from an image and then sharpen the latent text. STR has become widely researched because of application areas that cover most everyday activities for both humans and their technological advancement (such as self-driven vehicles and Artificial intelligence gadgets). Existing approaches must fully address the complex problems that arise in the wild regarding text recognition and detection, such as blurred images, which is an aspect of the task for this study. Again, research has recommended using Dimensionality Reduction (DR) and Genetic Algorithm (GA) to instantiate text recognition. Hence, it has given rise to using GA for DR as recommended. Here, we develop two major DR models: the DR model using ICA for pre-processing of the dataset and the DR model using an Independent Component Analysis (ICA) – Enhanced Genetic Algorithm using the Bird Approach (BA-GA), making the ICA-BA-GA model for text deblurring in the wild, coupled with SVM, K-NN, and Ensemble for evaluating the models. The study uses Large-Scale Street View 2019 ICDAR Text (LSVT19), which has 20,000 test photos, 30,000 annotated training images, and 400,000 unlabeled or partially labeled training images. Evaluation parameters such as accuracy, precision, and f1-scores were used for benchmarking. For accuracy compared to the state of the art, ICA-BA-GA gives an impressive 99%. For the improved model (ICA-BA-GA) using the classifiers - with the ensemble, we get the best result (99.30%): K-NN 98.65% and SVM 94.01%. Further research could investigate a hybrid using Neural Networks. A different dataset, preferably one curated by scholars with various categories of blurriness, can be used. Keywords: Scene Text Recognition, Independent Component Analysis, ICDAR Text (LSVT19), Deblurring Text, Pattern Recognition Word Count: 291
dc.identifier.citationKate Turabia
dc.identifier.otherP.hD
dc.identifier.urihttps://repository.lcu.edu.ng/handle/123456789/804
dc.language.isoen
dc.publisherLead City University, Ibadan
dc.relation.ispartofseriesP.hD
dc.subjectScene Text Recognition
dc.subjectIndependent Component Analysis
dc.subjectICDAR Text (LSVT19)
dc.subjectDeblurring Text
dc.subjectPattern Recognition
dc.titleHybridized Dimensionality Reduction Model for Blurred Text Detection in Natural Scene Images
dc.typeThesis

Files

Original bundle
Now showing 1 - 2 of 2
No Thumbnail Available
Name:
Nwufoh Chinonyelum Vivian Full Thesis (1).pdf
Size:
26.39 MB
Format:
Adobe Portable Document Format
Download
No Thumbnail Available
Name:
Nwufoh, Chinonyelum Vivian Preliminary Pages (1).pdf
Size:
1.7 MB
Format:
Adobe Portable Document Format
Download
License bundle
Now showing 1 - 1 of 1
No Thumbnail Available
Name:
license.txt
Size:
1.71 KB
Format:
Item-specific license agreed to upon submission
Description:
Download

Collections

Department of Computer Science;