Efficient Salient Region Detection with Soft Image Abstraction

[:en]

Ming-Ming Cheng      Jonathan Warrell       Wen-Yan Lin        Shuai Zheng       Vibhav Vineet         Nigel Crook

Vision Group, Oxford Brookes University

Abstract

Detecting visually salient regions in images is one of the fundamental problems in computer vision. We propose a novel method to decompose an image into large scale perceptually homogeneous elements for efficient salient region detection, using a soft image abstraction representation. By considering both appearance similarity and spatial distribution of image pixels, the proposed representation abstracts out unnecessary image details, allowing the assignment of comparable saliency values across similar regions, and producing perceptually accurate salient region detection. We evaluate our salient region detection approach on the largest publicly available dataset with pixel accurate annotations. The experimental results show that the proposed method outperforms 18 alternate methods, reducing the mean absolute error by 25.2% compared to the previous best result, while being computationally more efficient.

Paper

1. Efficient Salient Region Detection with Soft Image Abstraction. Ming-Ming Cheng, Jonathan Warrell, Wen-Yan Lin, Shuai Zheng, Vibhav Vineet, Nigel Crook. ICCV 2013. [pdf][bib][latex]

Supplemental materials

• Results comparisons to 18 alternative methods for MSRA 1000 dataset in a 79M PDF.
• Our result saliency maps: 31MB ZIP, results for other methods (360M ZIP).
• Prototype software: 2M ZIP.
• C++ source code is available. It runs 90 fps at my computer (CPU: Intel(R) core (TM) i7 cup 970 @ 3.2 GHz).

Other closely related projects:

1. Salient object detection and segmentation

2. Group saliency[:zh]

Ming-Ming Cheng      Jonathan Warrell       Wen-Yan Lin        Shuai Zheng       Vibhav Vineet         Nigel Crook

Vision Group, Oxford Brookes University

Abstract

Detecting visually salient regions in images is one of the fundamental problems in computer vision. We propose a novel method to decompose an image into large scale perceptually homogeneous elements for efficient salient region detection, using a soft image abstraction representation. By considering both appearance similarity and spatial distribution of image pixels, the proposed representation abstracts out unnecessary image details, allowing the assignment of comparable saliency values across similar regions, and producing perceptually accurate salient region detection. We evaluate our salient region detection approach on the largest publicly available dataset with pixel accurate annotations. The experimental results show that the proposed method outperforms 18 alternate methods, reducing the mean absolute error by 25.2% compared to the previous best result, while being computationally more efficient.

Paper

1. Efficient Salient Region Detection with Soft Image Abstraction. Ming-Ming Cheng, Jonathan Warrell, Wen-Yan Lin, Shuai Zheng, Vibhav Vineet, Nigel Crook. ICCV 2013. [pdf][bib][latex]

Supplemental materials

• Results comparisons to 18 alternative methods for MSRA 1000 dataset in a 79M PDF.
• Our result saliency maps: 31MB ZIP, results for other methods (360M ZIP).
• Prototype software: 2M ZIP.
• C++ source code is available. It runs 90 fps at my computer (CPU: Intel(R) core (TM) i7 cup 970 @ 3.2 GHz).

Other closely related projects:

1. Salient object detection and segmentation

2. Group saliency[:]