This paper presents a discriminative regional feature integration (DRFI) approach for salient object detection. The method formulates saliency map computation as a regression problem, utilizing multi-level image segmentation and supervised learning to map regional feature vectors to saliency scores. The key contributions include: 1. **Discriminative Regional Feature Integration (DRFI)**: The approach learns a Random Forest regressor to directly map regional feature vectors to saliency scores, automatically integrating high-dimensional features and selecting discriminative ones. 2. **Performance**: The DRFI approach significantly outperforms state-of-the-art methods on six benchmark datasets, including MSRA-B, iCoSeg, ECSSD, DUT-OMRON, SED2, and ECSSD, with large margins in terms of AUC scores, PR curves, and ROC curves. 3. **Efficiency**: The method runs as fast as most existing algorithms, making it suitable for real-world applications. 4. **Generalization**: The learned regressor demonstrates good generalization ability, outperforming other methods on challenging cases where the pseudo-background assumption may not hold well. The paper also discusses the limitations of the approach, such as its potential failure in cluttered scenes, and provides a detailed analysis of the most important features used by the Random Forest regressor. Overall, the DRFI approach offers a robust and efficient solution for salient object detection.This paper presents a discriminative regional feature integration (DRFI) approach for salient object detection. The method formulates saliency map computation as a regression problem, utilizing multi-level image segmentation and supervised learning to map regional feature vectors to saliency scores. The key contributions include: 1. **Discriminative Regional Feature Integration (DRFI)**: The approach learns a Random Forest regressor to directly map regional feature vectors to saliency scores, automatically integrating high-dimensional features and selecting discriminative ones. 2. **Performance**: The DRFI approach significantly outperforms state-of-the-art methods on six benchmark datasets, including MSRA-B, iCoSeg, ECSSD, DUT-OMRON, SED2, and ECSSD, with large margins in terms of AUC scores, PR curves, and ROC curves. 3. **Efficiency**: The method runs as fast as most existing algorithms, making it suitable for real-world applications. 4. **Generalization**: The learned regressor demonstrates good generalization ability, outperforming other methods on challenging cases where the pseudo-background assumption may not hold well. The paper also discusses the limitations of the approach, such as its potential failure in cluttered scenes, and provides a detailed analysis of the most important features used by the Random Forest regressor. Overall, the DRFI approach offers a robust and efficient solution for salient object detection.