Saliency Map Estimation Using a Pixel-Pairwise-Based Unsupervised Markov Random Field Model

Abstract

This work presents a Bayesian statistical approach to the saliency map estimation problem. More specifically, we formalize the saliency map estimation issue in the fully automatic Markovian framework. The major and original contribution of the proposed Bayesian–Markov model resides in the exploitation of a pixel pairwise modeling and a likelihood model based on a parametric mixture of two different class-conditional likelihood distributions whose parameters are adaptively and previously estimated for each image. This allows us to adapt our saliency estimation model to the specific characteristics of each image of the dataset and to provide a nearly parameter-free—hence dataset-independent—unsupervised saliency map estimation procedure. In our case, the parameters of the likelihood model are all estimated under the principles of the iterative conditional estimation framework. Once the estimation step is completed, the MPM (maximum posterior marginal) solution of the saliency map (which we show as particularly suitable for this type of estimation), is then estimated by a stochastic sampling scheme approximating the posterior distribution (whose parameters were previously estimated). This unsupervised data-driven Markovian framework overcomes the limitations of current ad hoc or supervised energy-based or Markovian models that often involve many parameters to adapt and that are finely tuned for each different benchmark database. Experimental results show that the proposed algorithm performs favorably against state-of-the-art methods and turns out to be particularly stable across a wide variety of benchmark datasets.

Salient Object Detection via Fusion of Multi-Visual Perception

“…The results of the proposed method for ECSSD and MASR10K datasets were compared with those of MSS [40], COV [41], GR [42], PCA [43], LDS [44], FCB [45], CNS [26], MDF [46], PPMRF [47], saliency bagging [2], MTL-ITSOD [48], PDLSO [7], and LC-SA [6] methods. These algorithms were selected to demonstrate the performance of our proposed method.…”

“…[40,41] can combat complex natural scenarios and [42,43], based on the prior theory, make results more robust. Others [26,46,47] are representative methods that emerged in recent years. All these methods utilize color and some prior theories to suppress background information and highlight salient objects.…”

“…Moreover, to further clarify the advantages of our approach relative to existing methodologies, we performed comparisons based on quantitative metrics (F-measure and MAE) with the MSS [40], COV [41], GR [42], PCA [43], LDS [44], FCB [45], CNS [26], MDF [46], PPMRF [47], saliency bagging [2], MTL-ITSOD [48], PDLSO [7], and LC-SA [6] methods using the ECSSD and MSRA10K datasets. Notably, due to the absence of results for the MTL-ITSOD and PDLSO algorithms for the MSRA10K dataset and the lack of results for the LC-SA algorithm for the ECSSD dataset, these instances are indicated in the table with a hyphen.…”

“…Analysis of the MAE and F-measure results from Table 2 reveals commendable performance of our proposed method for both the ECSSD and MSRA10K datasets. [41] 0.1941 0.9356 0.1765 0.9471 GR [42] 0.1522 0.9247 0.0814 0.9624 PCA [43] 0.1474 0.9311 0.0835 0.9668 LDS [44] 0.1824 0.9365 0.1537 0.9523 FCB [45] 0.1611 0.9369 0.1116 0.9611 CNS [26] 0.1469 0.9418 0.0947 0.9666 MDF [46] 0.1680 0.9388 0.1177 0.9620 PPMRF [47] 0.1466 0.9357 0.1264 0.9484 Saliency bagging [2] 0.1386 0.7698 0.0805 0.8838 MTL-ITSOD [48] 0.0047 0.9101 --PDLSO [7] 0.1210 0.8240 --redLC-SA [6] --0.1073 0.8198 Proposed 0.1302 0.9382 0.0783 0.9635…”