中央研究院 資訊科學研究所

In this paper, we propose a dictionary updating method, called the PMK-SVD method, and show numerically that it can stabilize the dictionary updating process, increase the convergence speed, and converge to a dictionary that outperforms the dictionary derived by the K-SVD method. The proposed method is based on the proximal point approach, which imposes a constraint on the distance of the dictionary modifications in the dictionary updating process. Specifically, we incorporate the approach into the well-known MOD and K-SVD dictionary updating algorithms and combine the results to obtain the PMK-SVD method. We analyze the complexity of the proposed method and compare it with that of the K-SVD method. The results of experiments demonstrate that our method outperforms K-SVD.

Morphological component analysis (MCA) for signal separation decomposes a signal into a superposition of morphological subcomponents, each of which is approximately sparse in a certain dictionary. Some of the dictionaries can also be modified to make them adaptive to local structure in images. We show that signal separation performance can be improved over the previous MCA approaches by replacing L1 norm optimization with “weighted” L1 norm optimization and replacing their dictionary adaptation with regularized dictionary adaptation. The weight on an atom for sparse coding is commonly derived from the corresponding coefficient’s value. In contrast, the weight of an atom in a dictionary for signal separation is derived from the mutual coherence between the atom and the atoms in the other dictionaries. The proposed solution for regularized dictionary adaptation is an extension of the K-SVD method, where the dictionary and “weighted” sparse coefficients are estimated simultaneously. We present a series of experiments demonstrating the significant performance improvement of the proposed algorithm over the previous approaches for signal separation.

High-dimensional local binary patterns [5] have been proved to be a useful feature for face recognition, which provides near-human performance in a widely used face verification benchmark. In this report, we review the technical aspect of this promising feature, and then we provide our implementation details of the feature. Finally, we show some experimental results using this feature on two public datasets, LFW and CACD.

In this paper, we have explored Google Glass, a modern wearable computing device, and developed a real-time system for the next decades of wearable computer. The system helps maximize the user interactions via wearable computer user interfaces, augmentation services and social network services. In our proposed system, we utilized Glass Development Kit/Mirror API, open source social network engine, and the augmentation of face recognition to analyze the human behavior/interaction while adapting to wearable computers in the coming era of Wearable Social Network. We expect it to alter how human behaves and interacts with each other in the near future, as the wearable computers are getting more popular in our daily life. Moreover, some issues of system performance, recognition techniques, user’s privacy and limitation are also discussed in this paper.