Visual Importance Identification of Natural Images Using Location-Based Feature Selection Saliency Map

Abstract

The proposed saliency map is called location-based feature selection saliency map (LBFSM). The research introduced a new method for identifying the image visual objects and region of unimportance. The saliency map uses Fourier transformation function for feature selection. The proposed method was applied over created natural images collected from different parts. The method’s efficiency was calculated based on objective and subjective quality assessment matrices such as processing time, precision and recall values and receiver operator characteristic (ROC) values. The quality assessment study showed the proposed saliency method efficiency in finding the local and global features from the image. The performance of the state-of-the-art saliency calculation method was experimented on the same natural image dataset. Five different saliency maps and their performance were compared and evaluated based on subjective and objective measures. Nine hundred (CRIST900) natural images were experimented using MATLAB R2015a, and their quality assessment was done using the same software platform. This research gives a conclusion that the result of processing time, receiver operator characteristic (ROC) curve, precision, and recall values provide good performance compared to the state-of-the art saliency map calculation methods.