Abstract

LEON GUERRA, Reynolds  and  GARCIA REYES, Edel B. Saliency detection in images in the quaternion space. Rev cuba cienc informat [online]. 2018, vol.12, n.4, pp. 128-139. ISSN 2227-1899.

In this paper, we developed an algorithm that allows the saliency detection in images in the quaternion space. To obtain maps of saliency, a Gaussian pyramid is built to process images at different scales through two approaches (local and global) while conserving salient objects. Images are represented in the quaternion space where the relationship between RGB and HSV color spaces is conserved using feature. In the local approach (spatial domain), the images are divided into patches for the comparison between the module and the phase of quaternion and a new variant of Quaternion Local Binary Patterns descriptor is developed. In the global approach, the images are processed in the quaternion Fourier frequency space, where the module spectrum is transformed by a low pass filter and the information is reconstructed to obtain a global map. The final map of saliency is achieved by a weighted combination of local and global maps and after applying a function of center bia more a refinement in the final map. To verify the effectiveness of our proposed algorithm it is validated using the mean absolute error metric in the ECSSD-1000 data set and compared with other state of the art algorithms.

Keywords : combination; quaternions; saliency regions.

        · abstract in Spanish     · text in Spanish     · Spanish ( pdf )