Saliency estimation

Estimation of image saliency can be defined as the task of assigning different levels of visual relevance to different regions in a digital image.
Despite the clear advantage that would be gained from solving this task, there is no universally accepted definition on what makes an element “salient”, thus rendering saliency estimation particularly challenging.


This can be better seen by observing the above figure; while the main object of interest in the first image can be generally recognized as the butterfly itself, the other two examples present less obvious answers.
We conduct an investigation with the specific objective of having a universal concept of image saliency to naturally emerge from a large set of heterogeneously-annotated data.

Publications

1.

Multiscale fully convolutional network for image saliency
(Simone Bianco, Marco Buzzelli, Raimondo Schettini) In Journal of Electronic Imaging, volume 27, pp. 27-27, 2018.

@article{bianco2018multiscale,
 author = {Bianco, Simone and Buzzelli, Marco and Schettini, Raimondo},
 year = {2018},
 pages = {27-27},
 title = {Multiscale fully convolutional network for image saliency},
 volume = {27},
 journal = {Journal of Electronic Imaging},
 url = {https://doi.org/10.1117/1.JEI.27.5.051221},
 pdf = {/download/2018a_CAMERA_Multiscale_fully_convolutional_network_for_image_saliency.pdf},
 doi = {10.1117/1.JEI.27.5.051221}}
2.

A Fully Convolutional Network for Salient Object Detection
(Simone Bianco, Marco Buzzelli, Raimondo Schettini) In International Conference on Image Analysis and Processing, pp. 82-92, 2017.

@inproceedings{bianco2017fully,
 author = {Bianco, Simone and Buzzelli, Marco and Schettini, Raimondo},
 year = {2017},
 pages = {82-92},
 title = {A Fully Convolutional Network for Salient Object Detection},
 organization = {Springer},
 booktitle = {International Conference on Image Analysis and Processing},
 pdf = {/download/2017b_CAMERA_A_Fully_Convolutional_Network_for_Salient_Object_Detection.pdf},
 doi = {10.1007/978-3-319-68548-9_8}}