Handcrafted Feature Pipelines for Image Classification: A Comparative Study of HOG, LBP, and FFT within a Transfer-Learning-Style Workflow

Emre Avuçlu; Filiz Sarı

doi:10.29002/asujse.1813481

EN TR

Handcrafted Feature Pipelines for Image Classification: A Comparative Study of HOG, LBP, and FFT within a Transfer-Learning-Style Workflow

Abstract

This study compares three handcrafted feature pipelines —HOG + histogram, LBP + statistics, and FFT + edge-density —within a transfer-learning-inspired workflow that employs an SVM classifier. Evaluation is performed on a balanced three-class image dataset (N = 1050; 350 images per class) using stratified 5-fold cross-validation and standard metrics (Accuracy, Precision, Recall, F1, ROC-AUC). Results indicate clear performance differences. The LBP-based pipeline achieves the highest overall accuracy (99.52%) and the most consistent class-wise behavior (macro-AUC ≈ 0.996), reflecting strong texture discrimination. The HOG-based pipeline attains robust performance (93.71% accuracy; macro-AUC ≈ 0.953), especially where edge and shape cues dominate. In contrast, the FFT-based pipeline is less effective overall (76.95% accuracy; macro-AUC ≈ 0.827), with reduced separability for texture-complex or low-contrast images. ROC-AUC analyses corroborate these findings across all classes, confirming the superiority of LBP features in this setting. Collectively, the results clarify when texture-centric, edge-centric, or frequency-centric descriptors are most advantageous and provide empirical guidance for feature selection in transfer-learning-style image classification pipelines, particularly under computational or data constraints.

Keywords

Görüntü Sınıflandırmada El Yapımı Öznitelik Hatları: Transfer öğrenmesi tarzı iş akışında HOG, LBP ve FFT’nin karşılaştırmalı analizi

Öz

Bu çalışma, SVM sınıflandırıcı kullanan transfer-öğrenmesi esinli bir iş akışı içinde üç el yapımı öznitelik hattını—HOG + histogram, LBP + istatistik ve FFT + kenar yoğunluğu—karşılaştırmaktadır. Değerlendirme, dengeli üç sınıflı bir görüntü veri kümesi üzerinde (N = 1050; sınıf başına 350 görüntü) katmanlı (stratified) 5 katlı çapraz doğrulama ve standart metrikler (Doğruluk, Kesinlik, Duyarlılık, F1, ROC-AUC) kullanılarak gerçekleştirilmiştir. Bulgular, belirgin performans farklarına işaret etmektedir. LBP tabanlı hat, en yüksek genel doğruluğa (%99,52) ve en tutarlı sınıf-bazlı davranışa (makro-AUC ≈ 0,996) ulaşarak güçlü doku ayrımlaşmasını yansıtmaktadır. HOG tabanlı hat, özellikle kenar ve şekil ipuçlarının baskın olduğu durumlarda, sağlam bir performans sergilemektedir (%93,71 doğruluk; makro-AUC ≈ 0,953). Buna karşılık, FFT tabanlı hat genel olarak daha düşük etkililik göstermektedir (%76,95 doğruluk; makro-AUC ≈ 0,827) ve karmaşık dokuya ya da düşük kontrasta sahip görüntülerde ayrıştırma gücü azalmaktadır. ROC-AUC analizleri bu bulguları tüm sınıflar için doğrulamakta ve bu bağlamda LBP özniteliklerinin üstünlüğünü teyit etmektedir. Bir bütün olarak sonuçlar, doku-merkezli, kenar-merkezli ve frekans-merkezli betimleyicilerin hangi koşullarda daha avantajlı olduğunu netleştirmekte ve özellikle hesaplama ya da veri kısıtları altında transfer-öğrenmesi tarzı görüntü sınıflandırma hatlarında öznitelik seçimi için ampirik bir rehber sunmaktadır.

Anahtar Kelimeler

References

[1] Gholizade, M., Soltanizadeh, H., Rahmanimanesh, M., & Sana, S. S. (2025). A review of recent advances and strategies in transfer learning. International Journal of System Assurance Engineering and Management, 16(3), 1123–1162. https://doi.org/10.1007/s13198-024-02684-2
[2] Zhuang, F., Qi, Z., Duan, K., Xi, D., Zhu, Y., Zhu, H., Xiong, H., & He, Q. (2021). A Comprehensive Survey on Transfer Learning. Proceedings of the IEEE, 109(1), 43–76. https://doi.org/10.1109/jproc.2020.3004555
[3] Soundarya, B., & Poongodi, C. (2025). A novel hybrid feature fusion approach using handcrafted features with transfer learning model for enhanced skin cancer classification. Computers in Biology and Medicine, 190, 110104. https://doi.org/10.1016/j.compbiomed.2025.110104
[4] Tassiopoulou, S., & Koukiou, G. (2024). Fusing Ground-Penetrating Radar Images for Improving Image Characteristics Fidelity. Applied Sciences, 14(15), 6808. https://doi.org/10.3390/app14156808
[5] Mame, A. B., & Tapamo, J. R. (2023). Parameter optimization of histogram-based local descriptors for facial expression recognition. PeerJ Computer Science, 9, e1388. https://doi.org/10.7717/peerj-cs.1388
[6] Rey-Díaz, A., Martín-Fernández, I., San-Segundo, R., & Gil-Martín, M. (2024). Frequency Analysis and Transfer Learning Across Different Body Sensor Locations in Parkinson’s Disease Detection Using Inertial Signals. ECSA-11, 82(1), 32. https://doi.org/10.3390/ecsa-11-20507
[7] Tsalera, E., Papadakis, A., Samarakou, M., & Voyiatzis, I. (2022). Feature Extraction with Handcrafted Methods and Convolutional Neural Networks for Facial Emotion Recognition. Applied Sciences, 12(17), 8455. https://doi.org/10.3390/app12178455
[8] Jaruenpunyasak, J., & Duangsoithong, R. (2021). Empirical Analysis of Feature Reduction in Deep Learning and Conventional Methods for Foot Image Classification. IEEE Access, 9. https://doi.org/10.1109/access.2021.3069625

[9] Dalal, N., & Triggs, B. (2005). Histograms of oriented gradients for human detection. Proceedings. 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition, CVPR 2005, I. https://doi.org/10.1109/cvpr.2005.177
[10] Liu, L., Zhao, L., Long, Y., Kuang, G., & Fieguth, P. (2012). Extended local binary patterns for texture classification. Image and Vision Computing, 30(2), 86-99. https://doi.org/10.1016/j.imavis.2012.01.001
[11] Yan, W., & Dong, Y. (2024). Local Directional Difference and Relational Descriptor for Texture Classification. Mathematics, 12(21), 3432. https://doi.org/10.3390/math12213432
[12] Huang, D., Shan, C., Ardabilian, M., Wang, Y., & Chen, L. (2011). Local binary patterns and its application to facial image analysis: A survey. IEEE Transactions on Systems, Man and Cybernetics Part C: Applications and Reviews (Vol. 41, Issue 6). https://doi.org/10.1109/tsmcc.2011.2118750
[13] Alessio, S. M. (2016). Digital Signal Processing and Spectral Analysis for Scientists. Springer International Publishing. https://doi.org/10.1007/978-3-319-25468-5
[14] Novelli, P., Meanti, G., Buigues, P. J., Rosasco, L., Parrinello, M., Pontil, M., & Bonati, L. (2025). Fast and Fourier features for transfer learning of interatomic potentials. Npj Computational Materials, 11(1), 293. https://doi.org/10.1038/s41524-025-01779-z
[15] Avuçlu, E., Taşdemir, Ş., & Köklü, M. (2023). A new hybrid model for classification of corn using morphological properties. European Food Research and Technology, 249(3), 835–847. https://doi.org/10.1007/s00217-022-04181-x
[16] Avuçlu, E., & Köklü, M. (2025). Fast and Accurate Classification of Corn Varieties Using Deep Learning With Edge Detection Techniques. Journal of Food Science, 90(7), e70439. https://doi.org/10.1111/1750-3841.70439
[17] Yasin, E. T., Ropelewska, E., Kursun, R., Cinar, I., Taspinar, Y. S., Yasar, A., Mirjalili, S., & Koklu, M. (2025). Optimized feature selection using gray wolf and particle swarm algorithms for corn seed image classification. Journal of Food Composition and Analysis, 145, 107738. https://doi.org/10.1016/j.jfca.2025.107738

Details

Primary Language

English

Subjects

Signal Processing

Journal Section

Research Article

Authors

Emre Avuçlu
0000-0002-1622-9059
Türkiye

Filiz Sarı ^*
0000-0001-8462-175X
Türkiye

Publication Date

December 24, 2025

Submission Date

October 30, 2025

Acceptance Date

December 17, 2025

Published in Issue

Year 2025 Volume: 9 Number: 2

DOI

https://doi.org/10.29002/asujse.1813481

IZ

https://izlik.org/JA93SC67YY

Cite

RIS / Bibtex

APA

Avuçlu, E., & Sarı, F. (2025). Handcrafted Feature Pipelines for Image Classification: A Comparative Study of HOG, LBP, and FFT within a Transfer-Learning-Style Workflow. Aksaray University Journal of Science and Engineering, 9(2), 95-99. https://doi.org/10.29002/asujse.1813481