Salient Object Detection via Objectness Measure

R. Sai Srivatsa, R. Venkatesh Babu

Abstract

Salient object detection has become an important task in many image processing applications. The existing approaches exploit background prior and contrast prior to attain state of the art results. In this paper, instead of using background cues, we estimate the foreground regions in an image using objectness proposals and utilize it to obtain smooth and accurate saliency maps. We propose a novel saliency measure called 'foreground connectivity' which determines how tightly a pixel or a region is connected to the estimated foreground. We use the values assigned by this measure as foreground weights and integrate these in an optimization framework to obtain the final saliency maps. We extensively evaluate the proposed approach on two benchmark databases and demonstrate that the results obtained are better than the existing state of the art approaches.

Introduction

Salient object detection aims to identify the most visually prominent objects in an image. Traditional methods typically rely on two key assumptions: the Contrast Prior (salient objects differ significantly from their surroundings) and the Boundary Prior (image edges are likely background). However, the Boundary Prior is often fragile; if a salient object touches the edge of the frame, these algorithms may incorrectly suppress parts of the object.

This paper introduces a "foreground-first" approach. By leveraging objectness proposals and a novel "foreground connectivity" metric, the authors produce smoother, more accurate saliency maps that avoid the pitfalls of boundary-based assumptions.

Methodology

The proposed pipeline consists of three primary stages designed to transition from raw image data to a refined saliency map.

  1. Objectness Map Generation: The authors utilize a simplified version of the BING (Binarized Normed Gradients) technique. Instead of a complex learned model, they apply an $8 \times 8$ Laplacian of Gaussian filter to identify regions with closed boundaries and center-surround patterns. This generates a pixel-wise objectness score, which is then grouped into SLIC superpixels.
  2. Foreground Connectivity: This core innovation quantifies how "connected" a region is to the estimated foreground. By constructing a graph where superpixels are nodes and edges represent color-space Euclidean distances, the algorithm calculates a connectivity score ($FG(R)$). Regions highly similar to the predicted foreground receive higher saliency weights.
  3. Saliency Optimization: To ensure spatial smoothness and filter out noise from initial estimates, the foreground weights are integrated into a principled optimization framework. This cost-minimization step ensures the entire object is highlighted uniformly.
The saliency detection pipeline: (a) input image, (b) foreground estimate, (c) objectness map, (d) foreground connectivity, and (e) final optimized saliency map.

Experimental Evaluation

The method was evaluated on two benchmark datasets: MSRA-1000 (1,000 images with simple backgrounds) and CSSD (200 images with complex, structured backgrounds).

The proposed approach outperformed four state-of-the-art techniques (GS, SF, MR, and SO) in terms of Mean Absolute Error (MAE):

  • MSRA-1000: Achieved an MAE of 0.064 (compared to the next best 0.068).
  • CSSD: Achieved an MAE of 0.132 (compared to the next best 0.136).

Qualitatively, the method produces cleaner masks that cover the entire object rather than just highlighting high-contrast edges. Furthermore, the algorithm is computationally efficient, processing images in approximately 0.27 seconds on a standard i5 CPU.

Conclusion

By prioritizing foreground estimation over background assumptions, this method effectively handles objects that touch image boundaries. While the current use of rectangular objectness proposals introduces some background noise, the integration of foreground connectivity and optimization results in highly accurate saliency maps. Future work may explore non-rectangular proposals and more sophisticated connectivity measures to further improve performance in extremely cluttered scenes.

Summary updated

Abstract

Salient object detection has become an important task in many image processing applications. The existing approaches exploit background prior and contrast prior to attain state of the art results. In this paper, instead of using background cues, we estimate the foreground regions in an image using objectness proposals and utilize it to obtain smooth and accurate saliency maps. We propose a novel saliency measure called `foreground connectivity' which determines how tightly a pixel or a region is connected to the estimated foreground. We use the values assigned by this measure as foreground weights and integrate these in an optimization framework to obtain the final saliency maps. We extensively evaluate the proposed approach on two benchmark databases and demonstrate that the results obtained are better than the existing state of the art approaches.
BibTeX
@inproceedings{srivatsa2015salient,
  author = {R. Sai Srivatsa and R. Venkatesh Babu},
  title = {{Salient Object Detection via Objectness Measure}},
  booktitle = {ICIP},
  year = {2015},
  url = {https://openreview.net/forum?id=2GytUAJWOGx},
}
Figure