Content-based blur image retrieval using quaternion approach and frequency adder LBP (Q2326125)

Based on a thorough review of the recently developed various CBIR techniques using local and global image features, the authors propose a novel descriptor -- frequency-adder-based local binary pattern (FALBP) -- based on quaternionic representation of images. The final representation is obtained on the basis of concatenation of results of applying the histogram approach on the extracted FALBPs. A no-reference image blur metric based on the cumulative probability of blur detection (CPBD) is used in the classification phase. Weighing values are assigned to each feature to improve the retrieval performance, where the weights are determined using a multi-objective optimization problem (MOP). The proposed technique is compared with other techniques -- MDLBP [\textit{S. R. Dubey} et al., IEEE Trans. Image Process. 25, No. 9, 4018--4032 (2016; Zbl 1408.94149)], CST [\textit{R. Lan} and \textit{Y. Zhou}, ``Medical image retrieval via histogram of compressed scattering coefficients'', IEEE J. Biomed. Health Inform. 21, No. 5, 1338--1346 (2017; \url{doi:10.1109/JBHI.2016.2623840})], QWLD [\textit{R. Lan} et al., ``Quaternionic Weber local descriptor of color images'', IEEE Trans. Circuits Syst. Video Tech. 27, No. 2, 261--274 (2017; \url{doi:10.1109/TCSVT.2015.2492839})] -- on the modified datasets Wang (Corel-1k) and KTH-TIPS (for each original image in the modified datasets four blur categories are added). The results show a higher precision rate of the presented scheme.