2025 Vol.8 Oct.N05

• Title: A case study on traditional and deep learning image retrieval
• Name: Shenghao Luo
• Company: University of Illinois - Urbana-Champaign,Champaign, IL 61820-5711
• Abstract:

  This paper explores the performance and efficiency trade-offs between traditional and deep learning–based methods in content-based image retrieval (CBIR). We compare color histogram features with deep ResNet-50 embeddings, both in full (2048-D) and PCA-compressed (96-D) forms, across standard CBIR tasks and under real-world image distortions. Using the UKBench dataset, we evaluate retrieval accuracy using mAP, MRR, and P@K metrics. Results show that deep features significantly outperform color histograms, while PCA compression retains most retrieval quality with greatly reduced storage and computation. However, under challenging conditions like cropping and blur, PCA features can degrade sharply. We also assess the robustness of both representations to common image augmentations. Our findings highlight the conditions under which PCA offers practical advantages, and when full-dimensional deep embeddings remain essential for high-precision CBIR.

  
  

• Keyword: Content-Based Image Retrieval; Machine-Learning Based Image Retrieval; Deep Learning Based Image Retrieval; Feature Representation
• DOI: 10.12250/jpciams2025091003
• Citation form: Shenghao Luo.A case study on traditional and deep learning image retrieval[J]. Computer Informatization and Mechanical System,2025,Vol.8,pp.

 [1] Mansoori, N. S., Nejati, M., Razzaghi, P., & Samavi, S. (2013, May). Bag of visual words approach for image retrieval using color information. In 2013 21st Iranian Conference on Electrical Engineering (ICEE) (pp. 1-6). IEEE.

 [2] Ma, H., Zhu, J., Lyu, M. R. T., & King, I. (2010). Bridging the semantic gap between image contents and tags. IEEE Transactions on Multimedia, 12(5), 462-473.

 [3] Koonce, B. (2021). ResNet 50. In Convolutional neural networks with swift for tensorflow: image recognition and dataset categorization (pp. 63-72). Berkeley, CA: Apress.

 [4] Rui, Y., Huang, T. S., & Chang, S. F. (1999). Image retrieval: Current techniques, promising directions, and open issues. Journal of visual communication and image representation, 10(1), 39-62.

 [5] Wei, J., Peng, B., Lee, X., & Palpanas, T. (2024). Det-lsh: a locality-sensitive hashing scheme with dynamic encoding tree for approximate nearest neighbor search. arXiv preprint arXiv:2406.10938.

 [6] Jing, Y., & Baluja, S. (2008). Visualrank: Applying pagerank to large-scale image search. IEEE Transactions on Pattern Analysis and Machine Intelligence, 30(11), 1877-1890.

 [7] Partio, M., Cramariuc, B., Gabbouj, M., & Visa, A. (2002, October). Rock texture retrieval using gray level co-occurrence matrix. In Proc. of 5th Nordic Signal Processing Symposium(Vol. 75, No. 1, pp. 511-524).

 [8] Wan, J., Wang, D., Hoi, S. C. H., Wu, P., Zhu, J., Zhang, Y., & Li, J. (2014, November). Deep learning for content-based image retrieval: A comprehensive study. In Proceedings of the 22nd ACM international conference on Multimedia (pp. 157-166).

 [9] Hasan, B. M. S., & Abdulazeez, A. M. (2021). A review of principal component analysis algorithm for dimensionality reduction. Journal of Soft Computing and Data Mining, 2(1), 20-30.

 [10] Xia, P., Zhang, L., & Li, F. (2015). Learning similarity with cosine similarity ensemble. Information sciences, 307, 39-52.

 [11] Targ, S., Almeida, D., & Lyman, K. (2016). Resnet in resnet: Generalizing residual architectures. arXiv preprint arXiv:1603.08029.

 [12] Kotei, E., & Thirunavukarasu, R. (2023). A systematic review of transformer-based pre-trained language models through self-supervised learning. Information, 14(3), 187.