Content-based remote sensing image retrieval using multi-scale local ternary pattern

This paper discusses a content based remote sensing image retrieval technique using multi-scale, patch-based local ternary pattern and fisher vector encoding. The technique downsamples an image in three scales, each downsampled image utilizes local ternary pattern to obtain upper and lower texture images, and divides them into dense patches to build a final histogram representation. This representation is then encoded into normalized fisher vectors. To this end, we focus on two standard remote sensing datasets namely 20-class satellite remote sensing dataset and 21-class land-cover dataset. Visual and quantitative results signify high precision for the proposed technique with an improvement of approximately 5.71% and 6.57% for 21-class land-cover and 20-class satellite remote sensing datasets respectively. (C) 2020 Elsevier Inc. All rights reserved.