Analysis of the influence of human faces for the estimation of salience in omnidirectional images

In this contribution a study dedicated to understanding the influence of the presence of human faces in a 360 degrees image on human perception is presented. Extensive research on saliency estimation in 2D images has shown that the presence of faces attracts human attention. Following these studies, recent 2D image quality assessment methods exploit face detection systems in their models. The application of these concepts to 360 degrees image is not straightforward. Furthermore, existing literature lacks a comparative study between the performance of face detection algorithms on various types of images (2D, fisheye, and omnidirectional) and how detected faces affect the procedure of saliency estimation. In this direction, we analyze the importance of person faces in a scene, by performing a set of subjective tests. From the performed analysis, it results that, even in 360 degrees images, human faces represent an important factor for image saliency. However, giving equal importance to all the detected faces does not lead to a better saliency estimation. Therefore, in this work, a study on the possible use of the face detector in estimating the salience of the 360 degrees image is performed.