The directional monogenic binary pattern (DMBP), a new texture descriptor to improve the description of the structure and orientation of the texture in an image, which in turn will improve image retrieval results, is proposed in this paper.

The new descriptor is based on monogenic signal representation (see Felsberg and Sommer [1]), and it automatically fuses the amplitude, orientation, phase, and information of the central pixel of the monogenic signal describing the image features. The method builds up a feature map for each characteristic (amplitude, orientation, and phase). The feature map includes ten bits to describe the information of the central pixel, which can describe more patterns than in the previous methods (using less than ten bits for describing the feature). Therefore, the information within the image is more accurately extracted and represented for use in the retrieval. The authors show improved accuracy during image retrieval when their method is used in different queries on two image collections, Corel-1000 and Corel-10000.

The algorithm introduced here represents the texture direction information in an image in a better way than the existing local binary pattern (LBP) algorithms. However, the authors do not provide a time comparison between their method and the LBPs, so this new texture-based algorithm might be slower and therefore unsuitable for online image retrieval tasks. At the same time, the method shows good improvement, so it could have good potential for an image retrieval system. To fully understand the content of this paper, readers need to have basic image processing knowledge on such topics as wavelets and Gabor filters, and some texture processing literature knowledge.