Mathematical Problems in Engineering / 2019 / Article / Tab 5

Review Article

Content-Based Image Retrieval and Feature Extraction: A Comprehensive Review

Table 5

A summary of the performance of deep-learning-based approaches for CBIR.

AuthorsDatasetsPurposeModelAccuracy

Krizhevsky et al. [87]ILSVRC-2010 and ILSVRC-2012Image classificationCNN37.50% top-1 and 17.00% top-5 error rate on ILSVRC-2010 and 15.3% top-5 error rate on ILSVRC-2012
Sun et al. [88]LFW (Labeled Face in the Wild)Face verificationConvNets DeepID97.45% accuracy
Karpathy and Fei-Fei [89]Flickr8K, Flickr30 K and MSCOCOGeneration of descriptions of image regionsCNN and multimodal RNNEncouraging results
Li et al. [90]MIRFlickr and NUS-WIDESocial image understandingDCEThe performance of CBIR 0.512 on MIRFlickr and 0.632 NUS-WID with k = 1000
Kondylidis et al. [82]INRIA Holidays, Oxford 5k, Paris 6k, UK BenchContent-based image retrievalCNN based tf-idfImproved results
Shi et al. [83]5356 skeletal muscle and 2176 lung cancer images with four types of diseasesHistopathology image classification and retrievalPDRH algorithm97.49% classification accuracy and MAP (97.49% and 97.33%)