Image Retrieval Using Low Level and Local Features Contents: A Comprehensive Review

Table 2

Summary of texture feature-based image retrieval techniques.

Year	Method	Similarity measure	Dataset	Performance measure (%)
1997	Subblock DCT DC and AC coefficients [31]	Modified Euclidean distance	200 images of woods, flowers, and sky with mountains	Average retrieval rate (ARR): 82
2017	Multiresolution RGB images, feature vector generated with DCT DC coefficients, and statistical features from the group of AC coefficients [32]	Euclidean distance	Corel-1K	Precision: 87.50
				Recall: 17.50
				F score: 29.16
			GHIM-10K	Precision: 82.50
				Recall: 3.30
				F score: 6.35
2013	Quantized histogram statistical texture features generation using DCT with DC and first 3 AC components [33]	Euclidean distance	Corel-1K	Precision: 80
				Recall: 81
				F score: 80
2000	Wavelet-based salient points, color moments, and Gabor moments of salient points [24]	Mean square error	COREL	Retrieval accuracy: 83.2
2017	LTrP and DWT with the artificial neural network [34]	Artificial neural network	Corel-1K	ARR: 97.9
			Corel-5k	ARR: 87.42
			Corel-10K	ARR: 74.13
2008	Curvelet transform with low-order statistics [25]	L2 distance	Brodatz texture database	(ARR: 79.54
2012	Ranklet transform and color moments with K-means clustering [35]	Euclidean distance	Wang dataset	Precision: 78.86
2012	Second-order local tetra patterns using vertical and horizontal derivatives of pixels direction [22]	Modified Canberra	Corel-1000	Precision: 75.9
				Recall: 48.7
			Brodatz texture database	Recall: 85.3
			MIT VisTex	Recall: 90.02
2012	Edge directional information feature based on local extrema [28]	Modified Canberra	Corel-5000	Precision: 48.8
				Recall: 21.1
			Corel-10K	Precision: 40.0
				Recall: 15.7
			Corel-1000	Precision:74.8
				Recall: 49.16
			Brodatz texture	Recall: 82.68