Suitable features for visual saliency computation in monochrome images

Saliency computation must consider usage of suitable combination of features along with an appropriate model to deliver desired performance. In this paper, important features for visual saliency computation are identified by analysing the performance of all possible combinations of a set of common candidate features. Combination of intensity and orientation is found to be suitable for monochrome images. Principal component analysis (PCA) on selected features is also performed. Usage of only the first principal component of selected features achieves good performance and reduces feature channels to save computational resources to a great extent.