Learning to Extract Focused Objects from Low DOF Images
This paper proposes an approach to extract focused objects (i.e., attention objects) from Low depth of field images. To recognize the focused object, we decompose the image into multiple regions, which are described by using three types of visual descriptors. Each descriptor is extracted from a representation of some aspects of local appearance, e.g., a spatially localized texture, color, or geometrical property. Therefore, the focus detection of a region can be achieved by the classification of extracted visual descriptors based on a binary classifier. We employ a boosting algorithm to learn the classifier with a cascade of decision structure. Given a test image, initial segmentation can be achieved using obtained classification results. Finally, we apply a post-processing technique to improve the results by incorporating region grouping and pixel-level segmentation. Experimental evaluation on a number of images demonstrates the performance advantages of the proposed method, when compared with state-of-the-art methods.
Hongliang Li, and King Ngi Ngan, "Learning to Extract Focused Object from Low DOF Image," IEEE Transactions on Circuits and Systems for Video Technology, vol. 21, no. 11, pp. 1571 - 1580, 2011.  [PDF]
In this work, we are especially interested in performing segmentation on focused objects, which usually correspond to the visual  attention objects in many photos, TV programs or film productions. The recognition of focused object provides an important and powerful cue for visual information processing including content-based coding, retrieval, browsing and surveillance. An example of focused object is illustrated in Fig. 1, which includes a white flower to be extracted. Fig. 1(b) shows the ground truth mask of the original image in Fig. 1(a). The goal of our work is to extract the focused flower from the original low-DOF image, which is described in Fig. 1(c).
Fig. 1. An example of focused object extraction. (a) Original image. (b) Ground truth mask. (c) Focused object of interest.
In this paper, we propose a method to segment focused objects from low-DOF images. Unlike existing methods that perform focus decision based on the measurement of the amount of high-frequency components from gray level image, our method learns to identify focused objects by using a set of color training images. Fig. 2 shows the framework of our proposed method, which consists of three parts, namely training, testing, and post-processing.
Fig. 2 Framework of our proposed method.
Training examples of focused object. Left: Original images. Right: Reference masks.

   Original images.       Ground truth                  Results  [7]                Results [3]                Our method

Fig. 3. Comparison results for test images, namely Bee, Racoon, Dragonfly, Red-flower, Frog and Judge from top to bottom, respectively. (a) Original images. (b) Ground truth masks. (c) Results for method [7]. (d) Results for method [3]. (e) Results for our method.


[7] C. Kim, ”Segmenting a low-depth-of-field image using morphological filters and region merging,” IEEE Trans. Image Processing, vol.14, no.10, pp.1503-1511, Oct. 2005.

[3] H. Li, and King N. Ngan, ”Unsupervised video segmentation with low depth of field”, IEEE Trans. Circuits and Systems for Video Technology, vol. 17, no.12, pp. 1742-1751, 2007.

The distribution of the segmentation errors for 117 test images is displayed in Fig. 4. This figure shows our algorithm peaking at a segmentation error of 0.0357, ahead of 0.1071 for the methods [3] and [7].
Fig. 4. Distribution of segmentation errors for 117 test images.
1. Ground truth Used for Comparison
89 images for training downloaded from here

117 test images and ground truth masks can be downloaded from here.

2. Segmentation Result by the proposed method can be downloaded from here
This work was partially supported by NSFC (No.60972109) and the Program for New Century Excellent Talents in University (NCET-08-0090).