Spectral clustering is a new graph and similarity based clustering algorithm. When the image is too big, it will take a long time to compute affinity matrix and its eigenvalues and eigenvectors. In order to improve the convergent speed of spectral clustering, a twostage texture segmentation algorithm is proposed in this paper. First, an improved watershed algorithm is used to perform pre-segmentation and then multiway spectral clustering with eigenvalue-scaled eigenvectors performs the final segmentation. This can reduce the runtime greatly and it is valuable to application with high time request. To verify the proposed algorithm, it is applied to texture image segmentation and the segmentation results are satisfying.

The performance of a perceptual hashing system, which is often measured by discrimination and robustness, is directly related to the features that the system extracts. In this letter, a new speech hashing scheme based on short-time stability is presented. The characteristic of natural speech that the principal components of linear prediction coefficients among neighboring frames tend to be very similar is utilized to generate the hash sequence. Experimental results demonstrate the effectiveness of the proposed scheme in terms of discrimination and robustness.

The visualization and interaction of the photolithography simulation result is useful and challenging. We introduce algorithms that include an automatic sorting algorithm, a transformation model and a dynamic loading algorithm to visualize the large-scale photolithography simulation data in three dimensions in real-time. The automatic sorting algorithm is presented to resort the data output by simulator. And the transformation model converts light intensity data into depth data for 3D modelling. Moreover, the dynamic loading algorithm realizes the out of core visualization. The experimental results demonstrate that our algorithms are effective for photolithography simulation data with 3D largescale interactive visualization in real-time.