Search

We greatly thank Zhang et al. (2008-this issue) for their comments on our recent paper (Zhang et al., 2007) regarding the tectonic setting of the Indosinian granitoids in the Songpan–Garze fold belt, as it gives us an opportunity to further clarify our viewpoints. The discrepancy between Zhang et al. (2008-this issue) and Zhang et al., 2007 H.-F. Zhang, R. Parrish, L. Zhang, W.-C. Xu, H.-L. Yuan, S. Gao and Q.-G. Crowley, A-type granite and adakitic magmatism association in Songpan–Garze fold belt, eastern Tibetan Plateau: implication for lithospheric delamination, Lithos 97 (2007), pp. 323–335. Article | PDF (1219 K) | View Record in Scopus | Cited By in Scopus (7)Zhang et al. (2007) is focused on the interpretation of the tectonic setting of the Indosinian A-type granite (211 ± 1 Ma) and adakitic magmatism (216–221 Ma) in the Songpan–Garze fold belt. We emphasize that these magmatic rocks formed in post-collisional tectonic setting, while Zhang et al. (2008-this issue) suggest that they were related to the northward subduction of the Jinsajian Tethys.

For input saturated Hammerstein systems, the two-step predictive control strategy is adopted. The first step calculates the desired intermediate variable applying unconstrained linear model and predictive control. The second step obtains the real-time control action by solving algebraic equation and desaturation. The case of immeasurable state is considered where the observer gain matrix is solved by Sylvester equation. The sufficient closed-loop stability condition is given and the designing and tuning algorithm for the domain of attraction is proposed. The theoretical results are validated by an example.

Binocular measurement system has a wide range of applications and becomes a hot research direction in recent years. Traditional feature extraction and matching methods limit its accuracy and robustness. In this paper, we propose a new method of binocular depth measurement based on Speed Up Robust Feature (SURF). Firstly, we extract SURF feature points from object original image as a standard feature database. Then, nearest neighbor method is used to find matching features in the right and left images. To eliminate error matching features, we propose to use Grubbs method to remove outliers of visual disparities. Finally, object depth is figured out by the binocular disparity. Experiments show its accuracy and robustness.

A new algorithm for face tracking is proposed. Color information provides an effective cue for face tracking due to its robustness to scaling, rotation and translation. But the main limitation of color cue is that it can be easily interrupted by the camouflage objects that have the same or similar color with the target face. In order to achieve robust face tracking performance, the texture feature is used for face description, which describes the detailed information of the face. The LBP (Local Binary Pattern) is a new way to extract the texture feature. Particle filter is a successful and powerful estimation method. This paper fuses the color cue and LBP cue under the framework of particle filter. The tracking results show a more robust face tracking performance compared with the method based on single cue.

For face detection in complex background pictures with single or multiple faces, this paper presents a new face-detection method. In YCbCr chrominance space, the distribution of the skin color is described as a Gaussian model and presented as a likelihood image. Otsu method is used to segment the skin pixels; At the same time, the skin regions are located in the KL skin color space, finally the two segmentation results under two chrominance spaces (YCbCr and KL) are fused at the pixel level. Experimental results show that this approach gets a more robust and accurate performance in detecting faces than in any single color space.

Human skin color distribution is relatively compact in a color space. That skin pixels in each frame are closed together as a ?dot cloud? is hypothesized, the shape evolution of ?dot cloud? in the color space from frame to frame is parameterized as translation, scaling and rotation. The linear combination of forecasts, which is consisted of 2-order Markov predictor and Wiener one step predictor, is proposed instead of single predictor to predict these parameters for the next frame which is to be segmented, and human skin biological feature is then adopted to remove camouflage noise. Extensive tests prove that this algorithm is quite sensitive to human color, and more accurate for human skin segmentation with Bayes classifier.

Gender classification is one of the most challenging problems in the field of pattern recognition. The pixel-based gray image recognition method is quite sensitive to illumination variation and has high dimensions for computation. PCA-based image feature recognition algorithm can reduce the image dimension, but it is only on the basis of optimal entropy to choose face features which neglects the different gender information between the male and female. In order to overcome the disturbance of non-essential information such as illumination variations and facial expression changing, a new algorithm is proposed in this paper. That is, the 2-D Gabor transform is used for extracting the face features; a new method is put forwards to decrease dimensions of Gabor transform output for speeding up SVM training; finally gender recognition is accomplished with SVM classifier. Good performance of gender classification test is achieved on a relative large scale and low-resolution face database.

In this paper, a new region-based approach to multi-resolution image fusion is presented. To make fusion decision better, a master image is selected from the source images. Then, it is segmented using the background subtraction based Mumford and Shah algorithm, which significantly improves the segmentation at quality and convergence speed. After that, the fusion rules on every region of the image are determined based on the best priority-first strategy. Moreover, the fusion on the whole image domain is replaced by using different rules on every segmented region of the image. The corresponding regions in other images are located by linear mapping methods. Some numerical results have been given and compared to the pixel-based fusion methods. Despite an increase in complexity, more intelligent fusion rules may be applied to remove side effects like reducing contrast and sensitive to error of registration.

Moving object segmentation in video sequence is the foundation of motion analysis and motion track. In this paper, a novel segmentation algorithm is proposed to solve illuminance abrupt change problem, which is based on MOGs and interframe gradient cross-correlation. Firstly, a primary foreground segmentation is obtained, where an adaptive MOGs (mixture of Gaussians) is established for each pixel's luminance. Secondly, luminance and chroma of each pixel are varying in a big scale followed abrupt illuminance change, which cause mismatch between a pixel's luminance and its MOGs, and misclassify a vast of background pixels as foreground. To adapt illuminance sudden variation, an improved method combining interframe gradient information is adopted to correct the initial segmentation. Finally, morphological methods are used to remove shadows and isolated noise pixels. Extensive experiments are performed with various video sequences, which prove that this method is robust and of high segmentation accuracy