Distortion-specific feature selection algorithm for universal blind image quality assessment

Nizami, Imran Fareed; Majid, Muhammad; Manzoor, Waleed; Khurshid, Khawar; Jeon, Byeungwoo

doi:10.1186/s13640-018-0392-5

EURASIP Journal on Image and Video Processing

Table 1 Overall performance comparison of proposed algorithm for cross-database validation

From: Distortion-specific feature selection algorithm for universal blind image quality assessment

IQA database for training	IQA database for testing	BIQA technique	All features				After feature selection algorithm
			SROCC	LCC	KCC	RMSE	SROCC	LCC	KCC	RMSE
CSIQ	LIVE	BRISQUE [41]	0.9311	0.9095	0.7677	0.2519	0.9650	0.9363	0.8173	0.2382
		BLIINDS II [39]	0.9365	0.8799	0.7434	0.2311	0.9791	0.9768	0.7884	0.2062
		GM-LOG [30]	0.9495	0.9542	0.7581	0.2006	0.9627	0.9642	0.7716	0.1981
		SSEQ [45]	0.7012	0.6827	0.5112	0.1940	0.7068	0.6868	0.5183	0.2188
		DIIVINE [38]	0.8475	0.8799	0.7722	0.6524	0.9713	0.9577	0.8379	0.4789
		CurveletQA [31]	0.6170	0.5828	0.4279	0.2184	0.6516	0.6303	0.4869	0.1923
	TID2013	BRISQUE [41]	0.8986	0.8997	0.7106	0.1791	0.9225	0.9293	0.7258	0.1580
		BLIINDS II [39]	0.8005	0.7671	0.6068	0.2338	0.8056	0.7951	0.6089	0.2187
		GM-LOG [30]	0.9051	0.9260	0.7000	0.1711	0.9074	0.9285	0.7055	0.1642
		SSEQ [45]	0.7728	0.7775	0.5650	0.1842	0.7819	0.7825	0.5692	0.1831
		DIIVINE [38]	0.8627	0.8610	0.8271	0.4781	0.9279	0.9669	0.8812	0.4286
		CurveletQA [31]	0.7763	0.7773	0.5708	0.1980	0.7869	0.7846	0.5810	0.1933
LIVE	CSIQ	BRISQUE [41]	0.8998	0.9075	0.7073	0.2244	0.9133	0.9116	0.7244	0.2181
		BLIINDS II [39]	0.9017	0.8928	0.6882	0.2634	0.9561	0.9754	0.7421	0.2293
		GM-LOG [30]	0.9108	0.9000	0.7043	0.2426	0.9433	0.9618	0.7368	0.2238
		SSEQ [45]	0.6884	0.6874	0.4875	0.2360	0.7266	0.7521	0.5283	0.2257
		DIIVINE [38]	0.8714	0.8431	0.6673	0.4746	0.8817	0.8953	0.7272	0.4344
		CurveletQA [31]	0.7122	0.7154	0.5284	0.2759	0.7350	0.7222	0.5440	0.2678
	TID2013	BRISQUE [41]	0.9072	0.8778	0.7185	0.2388	0.9100	0.8808	0.7235	0.2357
		BLIINDS II [39]	0.9056	0.8384	0.6940	0.2467	0.9253	0.8919	0.7182	0.2348
		GM-LOG [30]	0.9204	0.9106	0.7118	0.2234	0.9214	0.9180	0.7206	0.2188
		SSEQ [45]	0.8501	0.8481	0.6317	0.2123	0.8527	0.8489	0.6538	0.2085
		DIIVINE [38]	0.8672	0.8590	0.7209	0.2724	0.8876	0.8669	0.7264	0.2480
		CurveletQA [31]	0.8417	0.8342	0.6467	0.2256	0.8537	0.8352	0.6692	0.2321
TID2013	CSIQ	BRISQUE [41]	0.8665	0.8741	0.7196	0.2174	0.9238	0.9149	0.7617	0.1996
		BLIINDS II [39]	0.8747	0.8326	0.6535	0.2689	0.8946	0.8878	0.6719	0.2544
		GM-LOG [30]	0.8393	0.8366	0.6360	0.1989	0.8582	0.8602	0.6579	0.1834
		SSEQ [45]	0.7281	0.7168	0.5116	0.2236	0.7282	0.7372	0.5187	0.2165
		DIIVINE [38]	0.8481	0.7940	0.6175	0.2904	0.8790	0.8824	0.6672	0.2007
		CurveletQA [31]	0.7723	0.7604	0.5634	0.2202	0.7743	0.7665	0.5711	0.2153
	LIVE	BRISQUE [41]	0.9288	0.9256	0.7481	0.2536	0.9426	0.9272	0.7618	0.2342
		BLIINDS II [39]	0.9389	0.8917	0.7451	0.2279	0.9513	0.9228	0.7540	0.2071
		GM-LOG [30]	0.9336	0.9286	0.7349	0.2109	0.9582	0.9373	0.7596	0.2043
		SSEQ [45]	0.7046	0.6641	0.4958	0.2079	0.7268	0.7021	0.5210	0.1970
		DIIVINE [38]	0.8657	0.8243	0.6464	0.2919	0.9026	0.8628	0.7280	0.2332
		CurveletQA [31]	0.8121	0.7937	0.6122	0.2648	0.8209	0.8041	0.6252	0.2563

The italic values signify better performance when using all the features or proposed feature selection algorithm for a particular BIQA technique

Back to article page