Attention Prediction Using Partial Differential on Color Spaces
Zhiyong He, Song Sheng Lin · 2018
Human attention prediction has attracted significant interest in recent years due to its promising contributions to various computer vision applications. In this paper, we present a simple and effective framework for extracting meaningful features from the color space. The feature extraction is based on computing the second order partial differential for the input image. The features extracted from the RGB and LUV spaces are combined to form a saliency map of natural images. Based on the public dataset used in previous works, we compare our proposed algorithm with several competitive approaches presented in the literature and the results demonstrate that the proposed method is effective. We also apply the feature based on the LUV space to improve some of the previous approaches, and our experimental results show that the application is meaningful for enhancing the accuracy of human attention prediction.