Jinxu Tao

Continuous sign language recognition using level building based on fast hidden Markov model

HMM-based Level Building algorithm outperforms other methods.Performance rises when employing Grammar and sign length constraints.System runs faster by using a fast algorithm for HMM. Sign sequence segmentation and sign recognition are two main problems in continuous sign language recognition (CSLR) system. In recent years, dynamic time warping based Level Building (LB-DTW) algorithm has successfully dealt with both two challenges simultaneously. However, there still exists two crucial problems in LB-DTW: low recognition performance due to bad similarity function and offline due to high computation. In this paper, we use hidden Markov model (HMM) to calculate the similarity between the sign model and testing sequence, and a fast algorithm for computing the likelihood of HMM is proposed to reduce the computation complexity. Furthermore, grammar constraint and sign length constraint are employed to improve the recognition rate and a coarse segmentation method is proposed to provide the maximal level number. In experiments with a KINECT dataset of Chinese sign language containing 100 sentences composed of 5 signs each, the proposed method shows superior recognition performance and lower computation compared to other existing techniques.

An Isolated Sign Language Recognition System Using RGB-D Sensor with Sparse Coding

An isolated sign language recognition system is presented in this paper. A RGB-D sensor, Microsoft Kinect, is used for obtaining color stream and skeleton points from the depth stream. For a particular sign we extract a representative feature vector composed by hand trajectories and hand shapes. A sparse dictionary learning algorithm, Label Consistent K-SVD (LC-KSVD), is applied to obtain a discriminative dictionary. Based on that, we further develop a new classification approach to get better result. Our system is evaluated on 34 isolated Chinese sign words including one-handed signs and two-handed signs. Experimental results show the proposed system gets high recognition accuracy, of the reported 96.75%, and obtain an average accuracy of 92.36% for signer independent recognition.

Sign Language Recognition System Based on Weighted Hidden Markov Model

Sign language recognition (SLR) plays an important role in communication between deaf and hearing society. However, the recognition result is still worse for signer independent recognition. The reason is that there exists large variation between the signs from different subjects. In this paper, weighted hidden markov model (HMM) is proposed to deal with the variation. Unlike traditional HMM, WHMM assigns each sign samples with different weights. For the sign sample with big variation, the sample weight is big accordingly. Furthermore, we utilize Kinect to produce robust sign features to improve recognition rate. Our system is evaluated on one Chinese sign language dataset of 156 isolated sign words. Experimental result shows our proposed method outperforms other methods with a high recognition rate of 94.74%.

Deformation Corrected Blind Compressed Sensing (DC-BCS): A Novel Framework for Dynamic MRI reconstruction

Dynamic magnetic resonance imaging (MRI) is becoming vital important in modern clinical applications, and compared to other imaging methods such as B-ultrasound and CT, it has unique advantages. In this paper, we extend a deformation corrected blind compressed sensing (DC-BCS) method to reconstruct dynamic magnetic resonance data from under-sampled measurements. We introduce blind compressed sensing on the deformation corrected dynamic signals which avoids the need to know the sparsity basis in both the sampling and the recovery process. Then we will register the recovered images to the deformation corrected images and update them all in the each recovery iteration. The registration can be regard as a constraint to get a sparser representation and BCS techniques have been demonstrated to provide much better image reconstruction quality compared to compressed sensing methods that utilize a fixed sparsifying transform or dictionary. Combining them, we can achieve good performance. We jointly exploit the spatial and temporal sparsity actually and use variable splitting and alternative optimization to decouple the proposed complicated optimization problem to five easier subproblems. In order to avoid the risk of local convergence, we utilize effective continuation strategy. The results of experiment on the in-vivo dynamic myocardial perfusion MRI dataset show the proposed method achieves superior reconstruction quality, compared to the most state-of-the-art reconstruction methods.

Dynamic MRI reconstruction using low-rank and 3D sparsifying transform with separation of background and dynamic components

Dynamic magnetic resonance imaging (MRI) is an important auxiliary diagnostic method, and higher resolution of images is more conducive to the doctor to diagnose. In this paper, we extend a method which is referred to as robust principal component analysis (RPCA) to reconstruct dynamic magnetic resonance data from under-sampled measurements based on the low-rank plus sparse decomposition model. We consider the dynamic MRI as the sum of the background and the dynamic components, where the background is enforced low-rank by a non-convex function and a 3D sparsifying transform is used to enforce sparsity in the dynamic components. The proposed optimization problem is solved based on variable splitting and alternative optimization. The results of the in-vivo dynamic cardiac dataset show the proposed method achieves superior reconstruction quality, compared to the state-of-the-art reconstruction methods.

MR imaging reconstruction using a modified descent-type alternating direction method

In the magnetic resonance imaging (MRI) field, total variation (TV) which is the ℓ1 ‐norm of the gradient‐magnitude images (GMI) is widely used as the regularization in the compressive sensing (CS) based reconstruction algorithm. Based on the classic augmented Lagrangian multiplier method, we propose a modified descent‐type alternating direction method (ADM) for solving the TV regularized reconstruction problems in the following sense: an iteration result generated by the ADM is utilized to generate a descent direction; an appropriate step size along this descent direction is identified; and the penalty parameters are updated. The proposed algorithm effectively combines alternating direction technique with the descent‐type method. Extensive results demonstrate that the proposed algorithm, is competitive with, and often outperforms, other state‐of‐the‐art solvers in the field.