A cube-based saliency detection method using integrated visual and spatial features

Abstract

Purpose

The spatial feature is important for scene saliency detection. Scene-based visual saliency detection methods fail to incorporate 3D scene spatial aspects. This paper aims to propose a cube-based method to improve saliency detection through integrating visual and spatial features in 3D scenes.

Design/methodology/approach

In the presented approach, a multiscale cube pyramid is used to organize the 3D image scene and mesh model. Each 3D cube in this pyramid represents a space unit similar to a pixel in the image saliency model multiscale image pyramid. In each 3D cube color, intensity and orientation features are extracted from the image and a quantitative concave–convex descriptor is extracted from the 3D space. A Gaussian filter is then used on this pyramid of cubes with an extended center-surround difference introduced to compute the cube-based 3D scene saliency.

Findings

The precision-recall rate and receiver operating characteristic curve is used to evaluate the method and other state-of-art methods. The results show that the method used is better than traditional image-based methods, especially for 3D scenes.

Originality/value

This paper presents a method that improves the image-based visual saliency model.

Citation

Liu, T., Fang, Z., Mao, Q., Li, Q. and Zhang, X. (2016), "A cube-based saliency detection method using integrated visual and spatial features", Sensor Review, Vol. 36 No. 2, pp. 148-157. https://doi.org/10.1108/SR-07-2015-0110