Content-based image retrieval: A review of recent trends

Figures & data

Figure 1. General framework of the CBIR system

Figure 2. Classification of global features

Table 1. Summary of the literature on global feature-based methods

Ref.	Features	Feature extraction method	Dataset	Accuracy	Issues	Limitations
(Shrivastava & Tyagi, 2015)	Color Texture Shape	Color Histogram Gabor Filter Fourier Descriptor	Corel CIFAR	0.7690 0.8590	-Avoid linear dataset search.	-Color histogram does not provide any spatial information(Younus et al., 2015). - Fourier descriptor requires high computational cost.
(Younus et al., 2015)	Color Texture	Color Moment Color Histogram Wavelet Moment Co-occurrence Matrix	WANG	0.7352	- comparing query image with sub-set of dataset, because of employing clustering in offline stage. - clustering dataset using PSO and K-mean in offline stage.	- miss clustering cause irrelevant image retrieval. - Not including shape features in similarity measurement.
(Ponomarev et al., 2016)	Color Texture Shape	Color auto correlogram Gabor Transform Wavelet Transform	Corel Li Caltech101	0.8300 0.8800 0.7000	-considering color, texture and shape features in similarity measurement and acquiring high accuracy.	-high computational cost due to using Multiple features.
(Srivastava & Khare, 2017)	Texture Shape	Wavelet Transform/LBP Legendre Moments	Corel 1k Corel 5k Corel 10k Olivia 2688 GHIM10k	0.9995(7^thlevel) 0.5676(7^thlevel) 0.3537(7^thlevel) 0.9999(7^thlevel) 0.9172(7^thlevel)	- combining local and global features. - analyze images at multiple levels to capture missed details from other levels.	-Increased computational cost.
(Sajjad et al., 2018)	Color Texture	Quantized Color Histogram RLBP	Corel 1k Corel 10k ZB building	--0.8777	-Illumination and rotation invariant.	-No clear information about Results of Corel dataset.
(X. Zheng et al., 2016)	Color Texture	Projection Histogram Roberts Edge Detector	-	-	-Using main block to extract color features.	-No information about acquired results.
(Z. Zhao et al., 2016)	Color Texture Shape	CDE CLCM Hue Moments	Wang	0.9050	-Considering color, texture and shape features in similarity measurement. - High accuracy.	-accuracy depends on the nature of query image.
(Phadikar et al., 2018)	Color Texture	Color Histogram Color Moment MPEG-7 Edge descriptor	Corel 1k GHIM 10 K NUSWIND	0.9850	-Considering color and texture features in similarity measurement. - Using GA increased system accuracy. - Excellent accuracy.	-Using GA had an impact on computational cost.
(Pavithra & Sharmila, 2018)	Color Texture Shape	Color Moment LBP Canny Edge Detector	WANG Corel 5k Corel 10k	0.83225 0.6860 0.5998	-Avoid linear dataset search.	- The required running time depends on the number of images in the new sub-dataset.
(Pavithra & Sree Sharmila, 2019)	Color	Dominant Color Descriptor	Wang Corel-10 K OT-scene OxfordFlower	0.7534 0.4136 0.4352 0.3186	-introduce seed point selection to reduce the drawback of DCD.	-Do not bridge Semantic gap.
(Ashraf et al., 2018)	Color Shape	Color Histogram Canny Edge Histogram	WANG	0.7350	- Using artificial neural network (ANN) to learn the semantic class of the images	- Lack of spatial information. - No information about running time.
(Nazir et al., 2018)	Color Texture	Color Histogram DWT EDH	Corel	0.7350	- combining local and global features. - EHD are used to include the local distribution of the edges.	-Did not use any machine learning algorithm.
(Tadi Bani & Fekri-Ershad, 2019)	Color Texture	Quantized Color Histogram Gabor Filter GLCM	Simplicity	0.8294	-Extracting features in spatial and frequency domains. - Invariant to rotation and low sensitive to noise	-High run time.
(Rana et al., 2019)	Color Texture Shape	Color Moment Ranklet Transformation Invariant Moment	Simplicity Corel 5k Corel 10k Caltech101 MSR	0.6760 0.6744 0.6796 0.6450 0.6480	-Integration of non-parametric and parametric features.	-High computational cost due to high feature vector dimension.
(Thusnavis Bella & Vasuki, 2019)	Color Texture Shape	Color Moment GLCM Geometric Shape Feature	Corel 1k Corel 5k Corel 10k	0.8330 0.6690 0.5640	-Considering color, texture and shape features in similarity measurement.	- The retrieval time can be reduced if a suitable optimization algorithm is applied.
(Ashraf et al., 2020)	Color Texture	Color Moment DWT/Gabor Filter/CEDD	Corel1k Corel1.5k Corel5k GHIM20k	0.8750 0.8633 0.7983 0.7650	-High accuracy values.	-High computational cost due to high feature vector dimension.
(Alsmadi, 2020)	Color Texture Shape	GLCM DWT Canny Edge Histogram	Corel	0.9015	-Intensive and robust number of features.	-Criticality of cooling process. -Increase in calculation time.

Figure 3. Classification of local feature descriptors

Table 2. Summary of the literature for local feature-based methods

Ref.	Descriptor	Dataset	Accuracy	Issues	Limitations
(Montazer & Giveki, 2015)	SIFT LDP	Wang OT FP LSP	0.8850 0.9407 0.8860 0.8830	-High performance for images with objects.	- Length of feature vector is 3000. -Need to be improved for nature images.
(Sharif et al., 2019)	SIFT BRISK	Corel 1k Corel 1.5k Corel 5k Caltech 256	0.8439 0.7814 0.5737 0.4752	-Reduce semantic gap between high-level and low-level features.	-Not tested against large-scale unlabeled datasets.
(Yousuf et al., 2018)	SFIT LIOP	Corel 1k Corel 1.5k Caltech 256	0.8730 0.8520 0.3030	-Reduce semantic gap between high-level and low-level features. - Invariant to rotation, changing scale, illumination, and performing better in low contrast cases.	-High dimensional descriptor.
(Jabeen et al., 2018)	SURF FREAK	Corel 1k Corel 1.5k Caltech 256	0.8600 0.8320 0.3898	-Reduce semantic gap between high-level and low-level features.	-Did not offer any color information.
(Sarwar et al., 2019)	LBPV LIOP	WANG 1k WANG 1.5k Holiday	0.8958 0.7602 0.6923	-Reduce semantic gap between high-level and low-level features. - Using PCA to reduce dimensionality.	-Not tested against large-scale datasets. -Cannot be directly used for multispectral images.
(Mehmood et al., 2018)	HOG SURF	Corel 1k Corel 1.5k Corel 5k Caltech 256	0.8061 0.7628 0.6060 0.4630	-Offering more spatial information. -performs better in noise and low illumination situations.	-HOG Cannot be directly used for multispectral images.
(Baig et al., 2020)	CoHOG SURF	Corel 1k Corel 1.5k Sense 15 Caltech 256	0.8641 0.7726 0.8123 0.6839	-Using Relevance feedback to enhance accuracy.	-Not tested against large-scale datasets.
(Qin et al., 2019)	Harris SURF	Corel 1k	-	-Reduce search time.	- For large image dataset, the efficiency is not ideal.
(Sharmi et al., 2018)	-FAST	Medical dataset	-	-Using watermark-based protocol.	-FAST performs poorly in high level noise. FAST is dependent on threshold values.
(M. Zhao et al., 2016)	MTSD	Corel 1k Corel 5k Corel 10k Caltech 50	0.7928 0.6298 0.5195 0.4387	- Dimension of the proposed descriptor is 137	- Lack in describing the correlation.
(Raza et al., 2019)	STH	Corel 5k Corel 10k	0.6028 0.4803	- Correlating color information with texture orientation.	-Lack of intensity information.
(Raza et al., 2018)	CPV-THF	Corel 1k Corel 5k Corel 10k	0.8079 0.6390 0.5228	-Integrating semantic and content visual information.	- The dimension of the proposed feature vector is 242.
(song et al., 2018)	DTSD	Corel 1k Corel 5k UCID	0.6119 0.2019 0.3367	-Segmenting images to background and foreground.	- The dimension of the proposed feature vector is 692.
(Agarwal et al., 2019)	MCLTP	Corel 1k Corel 10k CMU-PIE STex MIT VisTex	0.8330 0.5690 0.8630 0.9810 0.4090	-Better retrieval performance.	- The dimension of the proposed feature vector is 3072.

Figure 4. Machine learning algorithms

Figure 5. SVM classifier

Figure 6. An example of CNN architecture

Table 3. Clarification of the main characteristics, limitations and examples for the main machine learning categories

Methods	Main characteristics	Limitations	Examples
Supervised Learning (Classification)	-requires the presence of labels. -predict and classify data to one of predefined classes.	- the size of the training set has a direct effect on the classification accuracy.	-SVM. -ANN.
Unsupervised Learning (Clustering)	-No need for labels. - learns the similarities and semantics between input data and generalize a model to handle unseen inputs.	-Over fitting. -Scalability. -Number of clusters affects the clustering algorithm performance.	-K-means. -K-means ++.
Deep Learning	-Supervised or Unsupervised. - Generate learning model	-Suffer from high computational cost. - Complex structure.	-CNN. -Deep Neural Network. -Deep Belief Network. -Boltzmann Machine.

Table 4. Summary of the performance of machine learning algorithms-based approaches for CBIR

Ref.	Machine learning algorithm	Dataset	Accuracy	Issues	Limitations
(Yousuf et al., 2018)	K-mean SVM	Corel 1k Corel 1.5k Caltech 256	0.8730 0.8520 0.3030	-Reduce semantic gap between high-level and low-level features.	High dimensional descriptor.
(Sarwar et al., 2019)	SVM	WANG 1k WANG 1.5k Holiday	0.8958 0.7602 0.6923	-Reduce semantic gap between high-level and low-level features. -Invariant to image rotation and monotonic intensity change. - Using PCA to reduce dimensionality.	-Not tested against large-scale datasets. -Cannot be directly used for multispectral images.
(Mehmood et al., 2018)	K-mean++ SVM	Corel 1k Corel 1.5k Corel 5k Caltech 256	0.8061 0.7628 0.6060 0.4630	-Offering more spatial information. -performs better in noise and low illumination situations.	-HOG Cannot be directly used for multispectral images.
(Ashraf et al., 2015)	ANN	Corel Coil Caltech 101	0.8200	-Retrieval based on the image core (main) object.	-Segmentation slow down the system.
(Alzu’bi et al., 2017)	CNN	Oxford 5k Oxford105k	0.9570 0.8860	-No need for annotations or labels. -Length of feature vector is 16. - Reduce required memory and run time.	- Accuracy decreases when using larger datasets.
(Tzelepi & Tefas, 2018)	CNN/Unsupervised CNN/Supervised	Paris 6k UKBench	0.9859 0.8347	- Reduce the dimension of the feature descriptor. - Retain spatial information.	- More retrieval time.
(Q. Zheng et al., 2019)	CNN	Caltech 101 Holiday Oxford	0.8850 0.9410 0.9620	-Using VGG-16. -Using similarity score.	-More enhancement is required in terms of speed of the training and testing stages. - More time to construct the gravitational field database
(Sezavar et al., 2019)	CNN	Corel ALOI MPEG-7	0.9559 0.9706 0.7749	-Reduce the computational cost.	- More retrieval time if not using sparse representation.

Table 5. Comparison of different CBIR techniques

Ref.	Feature extraction method	Accuracy	Strength of used method	Limitation
(Srivastava & Khare, 2017)	global	0.9999	- Provides directional information. - Provides spatial information. - Invariant to scale and rotation changes.	- Increases computation complexity. - Low accuracy when using large datasets. - DWT is sensitive to noise.
(Phadikar et al., 2018)	global	0.9850	-Considering color, texture and shape features in similarity measurement. - Using GA increased system accuracy. - Excellent accuracy.	- Criticality of the cooling process. -Using GA had an impact on computational time.
(Sarwar et al., 2019)	local	0.8958	-Reduce semantic gap between high-level and low-level features. -Invariant to image rotation and monotonic intensity change. - Using PCA to reduce dimensionality.	- Not tested against large datasets. - Loss in spectral and spatial information in multispectral images.
(Yousuf et al., 2018)	local	0.8730	-Reduce semantic gap between high-level and low-level features. - Invariant to rotation, changing scale, illumination, and performing better in low contrast cases.	- High dimensionality of SFIT.
(Tzelepi & Tefas, 2018)	CNN	0.9859	- Reduce the dimension of the feature descriptor. - Retain spatial information.	- More retrieval time.
(Sezavar et al., 2019)	CNN	0.9706	-Reduce the computational cost.	- More retrieval time if not using sparse representation.

Figure 7. Samples from corel, holiday, and brodatz image datasets

Table 6. Some widely used datasets in the CBIR domain

Dataset name	Number of images	Version	Number of classes	Number of images in each class
Corel (J. Z. Wang et al., 2000)	1000	Corel 1k	10	100
	1500	Corel 1.5k	15	100
	5000	Corel 5k	50	100
	10,000	Corel 10 K	10	100
WANG (J. Z. Wang et al., 2000)	1000	SIMPLicity	10	100
	10,000	WBIIS	10	1000
Holiday (Jain et al., 2011)	1491	Not- applicable	500	vary
Oliva (Oliva & Torralba, 2001)–(Das et al., 2016)	2688	QT-Scene	8	vary
CIFAR (Yang et al., 2018)	60,000	CIFAR-10	10	6000
Caltech (Fei-Fei et al., 2007)(Griffin et al., 2007)	More than 9000	Caltech 101	101	vary
	30,607	Caltech 256	256	80
GHIM (Srivastava & Khare, 2017) (Ashraf et al., 2020)	10,000	Not- applicable	20	500

Table 7. An example of two class confusion matrix

Predicted Classes
Actual Classes		A	B
	A	9	1
	B	2	8

Shrivastava, N., & Tyagi, V. (2015, August). An efficient technique for retrieval of color images in large databases. Computers & Electrical Engineering, 46, 314–327. https://doi.org/10.1016/j.compeleceng.2014.11.009

 Web of Science ®Google Scholar

Younus, Z. S., Mohamad, D., Saba, T., Alkawaz, H. M., Rehman, A., Al-Rodhaan, M., Al-Dhelaan, A. (2015, August). Content-based image retrieval using PSO and k-means clustering algorithm. Arabic Journal Geoscience, 8(8), 6211–6224. doi: 10.1007/s12517-014-1584-7

 Google Scholar

Ponomarev, A., Nalamwar, H. S., Babakov, I., Parkhi, C. S., & Buddhawar, G. (2016, February). Content-based image retrieval using color, texture and shape features. Key Engineering Materials, 685, 872–876. https://doi.org/10.4028/www.scientific.net/KEM.685.872

 Google Scholar

Srivastava, P., & Khare, A. (2017, January). Integration of wavelet transform, local binary patterns and moments for content-based image retrieval. Journal of Visual Communication and Image Representation, 42, 78–103. https://doi.org/10.1016/j.jvcir.2016.11.008

 Web of Science ®Google Scholar

Sajjad, M., Ullah, A., Ahmad, J., Abbas, N., Rho, S., & Baik, S. W. (2018, February). Integrating salient colors with rotational invariant texture features for image representation in retrieval systems. Multimedia Tools and Applications, 77(4), 4769–4789. https://doi.org/10.1007/s11042-017-5010-5

 Web of Science ®Google Scholar

Zheng, X., Tang, B., Gao, Z., Liu, E., & Luo, W. (2016, November). Study on image retrieval based on image texture and color statistical projection. Neurocomputing, 215, 217–224. https://doi.org/10.1016/j.neucom.2015.07.157

 Web of Science ®Google Scholar

Zhao, Z., Tian, Q., Sun, H., Jin, X., & Guo, J. (2016, January). Content based image retrieval scheme using color, texture and shape features. International Journal of Signal Processing Image Processing Pattern Recognition, 9(1), 203–212. doi: 10.14257/ijsip.2016.9.1.19

 Google Scholar

Phadikar, B. S., Phadikar, A., & Maity, G. K. (2018, May). Content-based image retrieval in DCT compressed domain with MPEG-7 edge descriptor and genetic algorithm. Pattern Analysis and Applications, 21(2), 469–489. https://doi.org/10.1007/s10044-016-0589-0

 Web of Science ®Google Scholar

Pavithra, L. K., & Sharmila, T. S. (2018, August). An efficient framework for image retrieval using color, texture and edge features. Computers & Electrical Engineering, 70, 580–593. https://doi.org/10.1016/j.compeleceng.2017.08.030

 Web of Science ®Google Scholar

Pavithra, L. K., & Sree Sharmila, T. (2019, December). An efficient seed points selection approach in dominant color descriptors (DCD). Cluster Computing, 22(4), 1225–1240. https://doi.org/10.1007/s10586-019-02907-3

 Web of Science ®Google Scholar

Ashraf, R., Ahmedm, M., Jabbar, S., Khalid, S., Ahmad, A., Din, S. & Jeon, G. (2018, March). Content based image retrieval by using color descriptor and discrete Wavelet transform. Journal of Medical Systems, 42 (3), 44. https://doi.org/10.1007/s10916-017-0880-7

 PubMed Web of Science ®Google Scholar

Nazir, A., Ashraf, R., Hamdani, T., & Ali, N., “Content based image retrieval system by using HSV color histogram, discrete wavelet transform and edge histogram descriptor,” in 2018 International Conference on Computing, Mathematics and Engineering Technologies (iCoMET), Sukkur, Pakistan, 2018, January, pp. 1–6.

 Google Scholar

Tadi Bani, N., & Fekri-Ershad, S. (2019, August). Content-based image retrieval based on combination of texture and colour information extracted in spatial and frequency domains. Electronic Library, 37(4), 650–666. https://doi.org/10.1108/EL-03-2019-0067

 Web of Science ®Google Scholar

Rana, S. P., Dey, M., & Siarry, P. (2019, January). Boosting content based image retrieval performance through integration of parametric & nonparametric approaches. Journal of Visual Communication and Image Representation, 58(3), 205–219. https://doi.org/10.1016/j.jvcir.2018.11.015

 Web of Science ®Google Scholar

Thusnavis Bella, M. I., & Vasuki, A. (2019, May). An efficient image retrieval framework using fused information feature. Computers & Electrical Engineering, 75, 46–60. https://doi.org/10.1016/j.compeleceng.2019.01.022

 Web of Science ®Google Scholar

Ashraf, R., Ahmed, M., Ahmad, U., Habib, M. A., Jabbar, S., & Naseer, K. (2020, April). MDCBIR-MF: Multimedia data for content-based image retrieval by using multiple features. Multimedia Tools and Applications, 79(13–14), 8553–8579. https://doi.org/10.1007/s11042-018-5961-1

 Web of Science ®Google Scholar

Alsmadi, M. K. (2020, April). Content-based image retrieval using color, shape and texture descriptors and features. Arabian Journal for Science and Engineering, 45(4), 3317–3330. https://doi.org/10.1007/s13369-020-04384-y

 Web of Science ®Google Scholar

Montazer, G. A., & Giveki, D. (2015, September). Content based image retrieval system using clustered scale invariant feature transforms. Optik (Stuttg), 126(18), 1695–1699. https://doi.org/10.1016/j.ijleo.2015.05.002

 Web of Science ®Google Scholar

Sharif, U., Mehmood, Z., Mahmood, T., Javid, M. A., Rehman, A., & Saba, T. (2019, August). Scene analysis and search using local features and support vector machine for effective content-based image retrieval. Artificial Intelligence Review, 52(2), 901–925. https://doi.org/10.1007/s10462-018-9636-0

 Web of Science ®Google Scholar

Yousuf, M., Mehmood, Z., Habib, H. A., Mahmood, T., Saba, T., Rehman, A., & Rashid, M. (2018). A novel technique based on visual words fusion analysis of sparse features for effective content-based image retrieval. Mathematical Problems in Engineering, 2018, 1–13. https://doi.org/10.1155/2018/2134395

 Web of Science ®Google Scholar

Jabeen, S., Mehmood, Z., Mahmood, T., Saba, T., Rehman, A., Mahmood, M. T., & Rubin, D. L. (2018, April). An effective content-based image retrieval technique for image visuals representation based on the bag-of-visual-words model. PLoS One, 13(4), e0194526. https://doi.org/10.1371/journal.pone.0194526

 PubMed Web of Science ®Google Scholar

Sarwar, A., Mehmood, Z., Saba, T., Qazi, K. A., Adnan, A., & Jamal, H. (2019, February). A novel method for content-based image retrieval to improve the effectiveness of the bag-of-words model using a support vector machine. Journal of Information Science, 45(1), 117–135. https://doi.org/10.1177/0165551518782825

 Web of Science ®Google Scholar

Mehmood, Z., Abbas, F., Mahmood, T., Javid, M. A., Rehman, A., & Nawaz, T. (2018, December). Content-based image retrieval based on visual words fusion versus features fusion of local and global features. Arabian Journal for Science and Engineering, 43(12), 7265–7284. https://doi.org/10.1007/s13369-018-3062-0

 Web of Science ®Google Scholar

Baig, F., Mehmood, Z., Rashid, M., Javid, M. A., Rehman, A., Saba, T., & Adnan, A. (2020, March). Boosting the performance of the BoVW model using SURF–CoHOG-based sparse features with relevance feedback for CBIR. Iranian Journal of Science and Technology, Transactions of Electrical Engineering, 44(1), 99–118. https://doi.org/10.1007/s40998-019-00237-z

 Web of Science ®Google Scholar

Qin, J., Li, H., Xiang, X., Tan, Y., Pan, W., Ma, W., & Xiong, N. N. (2019). An encrypted image retrieval method based on harris corner optimization and LSH in cloud computing. IEEE Access, 7, 24626–24633. https://doi.org/10.1109/ACCESS.2019.2894673

 Web of Science ®Google Scholar

Sharmi, N., Mohamed Shameem, P., & Parvathy, R. (2018). Content-based image retrieval using fast machine learning approach in cloud computing. In Communications in computer and information science (Vol. 837, pp. 434–444). Elsevier: Springer Singapore.

 Google Scholar

Zhao, M., Zhang, H., & Sun, J. (2016, July). A novel image retrieval method based on multi-trend structure descriptor. Journal of Visual Communication and Image Representation, 38, 73–81. https://doi.org/10.1016/j.jvcir.2016.02.016

 Web of Science ®Google Scholar

Raza, A., Nawaz, T., Dawood, H., & Dawood, H. (2019). Square texton histogram features for image retrieval. Multimedia Tools and Applications, 78(3), 2719–2746. https://doi.org/10.1007/s11042-018-5795-x

 Web of Science ®Google Scholar

Raza, A., Dawood, H., Dawood, H., Shabbir, S., Mehboob, R., & Banjar, A. (2018). Correlated primary visual texton histogram features for content base image retrieval. IEEE Access, 6, 46595–46616. https://doi.org/10.1109/ACCESS.2018.2866091

 Web of Science ®Google Scholar

Song, W., Zhang, Y., Liu, F., Chai, Z., Ding, F., Qian, X., & Park, S. C. (2018, April). Taking advantage of multi-regions-based diagonal texture structure descriptor for image retrieval. Expert Systems with Applications, 96, 347–357. https://doi.org/10.1016/j.eswa.2017.12.006

 Web of Science ®Google Scholar

Agarwal, M., Singhal, A., & Lall, B. (2019, November). Multi-channel local ternary pattern for content-based image retrieval. Pattern Analysis and Applications, 22(4), 1585–1596. https://doi.org/10.1007/s10044-019-00787-2

 Web of Science ®Google Scholar

Ashraf, R., Bashir, K., Irtaza, A., & Mahmood, M. (2015, May). Content based image retrieval using embedded neural networks with bandletized regions. Entropy, 17(6), 3552–3580. https://doi.org/10.3390/e17063552

 Web of Science ®Google Scholar

Alzu’bi, A., Amira, A., & Ramzan, N. (2017, August). Content-based image retrieval with compact deep convolutional features. Neurocomputing, 249, 95–105. https://doi.org/10.1016/j.neucom.2017.03.072

 Web of Science ®Google Scholar

Tzelepi, M., & Tefas, A. (2018, January). Deep convolutional learning for content based image retrieval. Neurocomputing, 275, 2467–2478. https://doi.org/10.1016/j.neucom.2017.11.022

 Web of Science ®Google Scholar

Zheng, Q., Tian, X., Yang, M., & Wang, H. (2019). Differential learning: A powerful tool for interactive content-based image retrieval. Engineering Letters, 27(1), 202–215. http://www.engineeringletters.com/issues_v27/issue_1/EL_27_1_23.pdf

 Web of Science ®Google Scholar

Sezavar, A., Farsi, H., & Mohamadzadeh, S. (2019, August). Content-based image retrieval by combining convolutional neural networks and sparse representation. Multimedia Tools and Applications, 78(15), 20895–20912. https://doi.org/10.1007/s11042-019-7321-1

 Web of Science ®Google Scholar

Wang, J. Z., Li, J., & Wiederholdy, G. (2000). SIMPLIcity: Semantics-sensitive Integrated Matching for Picture LIbraries. Lecture Notes in Computer Science, 1929 (9), 360–371. (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). DOI: 10.1007/3-540-40053-2_32

 Google Scholar

Jain, M., Jégou, H., & Gros, P., “Asymmetric hamming embedding,” in Proceedings of the 19th ACM international conference on Multimedia - MM ’11, New York, USA, 2011, no. May, p. 1441.

 Google Scholar

Oliva, A., & Torralba, A. (2001). Modeling the shape of the scene: A holistic representation of the spatial envelope. International Journal of Computer Vision, 42(3), 145–175. https://doi.org/10.1023/A:1011139631724

 Web of Science ®Google Scholar

Das, R., Thepade, S., Bhattacharya, S., & Ghosh, S. (2016). Retrieval architecture with classified query for content based image recognition. Applied Computer Intellegent Software Computers, 2016, 1–9. doi: 10.1155/2016/1861247

 Google Scholar

Yang, H.-F., Lin, K., & Chen, C.-S. (2018, February). Supervised learning of semantics-preserving hash via deep convolutional neural networks. IEEE Transactions on Pattern Analysis and Machine Intelligence, 40(2), 437–451. https://doi.org/10.1109/TPAMI.2017.2666812

 PubMed Web of Science ®Google Scholar

Fei-Fei, L., Fergus, R., & Perona, P. (2007, April). Learning generative visual models from few training examples: An incremental Bayesian approach tested on 101 object categories. Computer Vision and Image Understanding, 106(1), 59–70. https://doi.org/10.1016/j.cviu.2005.09.012

 Web of Science ®Google Scholar

Griffin, G., Holub, A., & Perona, P. (2007). Caltech-256 object category dataset. Caltech mimeo, 11(1), 20. https://authors.library.caltech.edu/7694/

 Google Scholar