Wavelet Transform Derived Dominant Descriptors for Rotation Invariant Image Retrieval

Abstract

In this paper, we introduce a new global level and geometrically invariant descriptor, coined as dominant edgels descriptor (DED). The proposed DED is integrated with geometrically invariant dominant local binary pattern (DRLBP) and dominant color descriptor (DCD) for photo, texture and histopathological image retrieval. The DRLBP and DED are exploited from multi-resolution domain and DCD is computed from RGB image. Experiments are conducted exhaustively on benchmark databases and it proves that proposed integration of geometrically invariant feature descriptors gives better results than the state-of-the-art methods.