An Improved Neural Network Cascade for Face Detection in Large Scene Surveillance

Face detection for security cameras monitoring large and crowded areas is very important for public safety. However, it is much more difficult than traditional face detection tasks. One reason is, in large areas like squares, stations and stadiums, faces captured by cameras are usually at a low resolution and thus miss many facial details. In this paper, we improve popular cascade algorithms by proposing a novel multi-resolution framework that utilizes parallel convolutional neural network cascades for detecting faces in large scene. This framework utilizes the face and head-with-shoulder information together to deal with the large area surveillance images. Comparing with popular cascade algorithms, our method outperforms them by a large margin.

Compactly Supported Tight and Sibling Frames Based on Generalized Bernstein Polynomials

We obtain a family of refinable functions based on generalized Bernstein polynomials to provide derived properties. The convergence of cascade algorithms associated with the new masks is proved, which guarantees the existence of refinable functions. Then, we analyze the symmetry, regularity, and approximation order of the refinable functions, which are of importance. Tight and sibling frames are constructed and interorthogonality of sibling frames is demonstrated. Finally, we give numerical examples to explicitly illustrate the construction of the proposed approach.

Study on error reconciliation in quantum key distribution

As one of the most important procedure in quantum key distribution system, the error reconciliation algorithm has drew many attentions. However, studies on the error reconciliation algorithm mainly focuses on the reconciliation efficiency. Since the ultimate goal of study on the error reconciliation is to find the most suitable algorithm for a quantum key distribution system and maximize the throughput rate of the whole system, the indicator of reconciliation efficiency is not full-scale enough to evaluate an error reconciliation algorithm. In this paper we propose a new evaluation scheme, including four direct indicators and one composite indicator to solve the problem. Following the new scheme, seven representative error reconciliation algorithms are simulated and compared thoroughly, i.e. BBBSS, the original Cascade and two improved Cascade algorithms, Winnow, and two LDPC based algorithms. Our works are very beneficial to the evaluation, comparison, selection and optimization of error reconciliation algorithms for a practical quantum key distribution system.

Convergence Rates of Cascade Algorithms with Infinitely Supported Masks

We investigate the solutions of refinement equations of the formwhere the function ϕ is in Lp(ℝs)(1 ≤ p ≤ ∞), a is an infinitely supported sequence on ℤs called a refinement mask, and M is an s × s integer matrix such that limn→1M–n = 0. Associated with the mask a and M is a linear operator Qa,M defined on Lp(ℝs) by Qa,Mϕ0 := Σα∈ℤsa(α)ϕ0(M · –α). Main results of this paper are related to the convergence rates of in Lp(ℝs) with mask a being infinitely supported. It is proved that under some appropriate conditions on the initial function ϕ0, converges in Lp(ℝs) with an exponential rate.

A CONSTRUCTION OF CONVERGENT CASCADE ALGORITHMS IN SOBOLEV SPACES

Cascade algorithms play an important role in wavelet analysis and computer graphics. The paper considers the convergence of cascade algorithms in Sobolev spaces. With the help of the factorization of matrix masks, we give a sufficient condition for the convergence. The condition is expressed in the time domain. More importantly, an algorithm for the construction of convergent cascade algorithms in Sobolev space starting from any matrix mask satisfying a mild condition is presented. Examples are given to illustrate our theorems.