Zhengyou Wang

Locating human eyes using edge and intensity information

In this paper, a new eye detection method is presented. The method consists of three steps: (1) extraction of binary edge image (BEI) based on the multi-resolution analysis of wavelet transform; (2) extraction of eye region and segments from BEI, and (3) eye localization using light dot or intensity information. An improved face region extraction algorithm and a light dot detection method are proposed to improve eye detection performance. Experimental results show that our approach can achieve a correct eye detection rate of 98.7% on 150 Bern images with variations in view and gaze direction and a rate of 96.6% on 564 AR images with different facial expressions and lighting conditions.

Modeling MPEG-4 VBR Video Traffic by Using ANFIS

Video traffic predicting and modeling are very important for compressed video transmission. The traditional method delineates the process by a rigid model with several parameters, which are difficult to estimate. In this paper, the MPEG-4 VBR (Variable Bit Rate) video traffic is modeled by the ANFIS (Adaptive Neuro-Fuzzy Inference System). Then, it is applied to modeling and predicting the MPEG-4 VBR video traffic. Simulations show the GoP (Group of Pictures) loss probabilities in actual video traffic are very close to those in ANFIS modeling traffic at the same experimental conditions and the prediction errors (1/SNR) are very small.

Evaluation of a Crossover Router Based QoS Mechanism in Fast Mobile IPv6 Networks

A “Crossover Router” (CR) based QoS provisioning mechanism under an enhanced “Fast Handovers for Mobile IPv6” (FMIPv6) architecture has been proposed to provide better performance for multimedia applications in Mobile IPv6 (MIPv6) networks. In the proposal we extend the FBU and HI messages to notify the QoS requirement of “Mobile Node” (MN). Advance reservations along the possible future-forwarding paths are performed only between CR and MN to reduce reservation hops and signaling delays. In this paper we present the detailed performance evaluation results of this scheme compared with “Mobile RSVP” (MRSVP) protocol, using the network simulator (NS-2). In the simulations RSVP signaling cost, total bandwidth requirements, RSVP signaling and data packet delay in different scenarios of both schemes were considered. Results show that the proposed scheme for QoS guarantee has lower RSVP signaling cost and delay, as well as less bandwidth requirements in comparison with MRSVP. Furthermore, fluctuation and peak values of data packet delays in different scenarios are much lower.

One Novel Subspace-based Method for MRS Signal Parameter Estimation

The method HTLS-PK that successfully incorporates prior knowledge into the classical method HTLS has been proven to greatly improve the performance of parameter estimation of overlapping peaks of MRS signal. Based on prior knowledge and decimation, a novel subspace-based parameter estimation method HTLSDStackPK is presented in this paper. The simulated results and relevant analysis show that HTLSDStackPK also has much better performance of parameter estimation than HTLS at high noise levels. Compared with HTLS-PK, HTLSDStackPK does not lead to better estimation, but the different is not high, and computational complexity of HTLSDStackPK is much lower than that of HTLS-PK

A new color image enhancement algorithm for camera-equipped mobile telephone

We present in this paper a color image enhancement algorithm with high speed, high quality, and low memory usage for the camera-equipped mobile telephone. To achieve this, we first introduce a propagation scheme that estimates and computes the statistical characteristics of an image, then we apply a piecewise nonlinear transformation in the R/G/B channels separately. If the value of a pixel is below minimum or above maximum, it can be ignored by using range stretching, then a narrow intensity range between minimum and maximum may be expanded to a much broader range defined by an exponent function, which consists of mean and variance of the pixel values, and its power is chosen by the trials. The optimal strategies for fast processing in the code style are also discussed. We demonstrate our color image enhancement system and compare it with other feasible techniques on the camera-equipped mobile telephone. The proposed algorithm can be carried out in about 2 seconds using the ARM7 processor and generally performs much better than traditional contrast enhancement methods with higher image quality.

Detecting All-Zero Coefficient Blocks Before Transformation and Quantization in H.264/AVC

In this paper a new early detection algorithm for all-zero coefficient blocks before integer transformation and quantization is proposed in very low bit rate video coding for H.264/AVC standard. The all-zero detecting threshold is very critical as a tradeoff between encoder complexity and image quality. So firstly a feasible threshold range is analyzed theoretically, then based on extensive experiments on typical video sequences from the statistical opinion, a better optimal threshold is discussed, which has a better balance than the theoretically optimal threshold between mis-justice and leak-out ratio in detecting all-zero coefficient blocks. In our proposed algorithm, the sum of absolute difference (SAD) of each motion prediction error in the inter block is used, and this SAD is a bypass obtained from motion estimation process, so no additional computation is required. The simulation results show that the proposed algorithm can reduce the computational load significantly, almost without video quality loss.

An efficient mobility management scheme for hierarchical mobile IPv6 networks

Performance evaluation shows that Hierarchical Mobile IPv6 (HMIPv6) cannot outperform standard Mobile IPv6 (MIPv6) in all scenarios. Thus, adaptive protocol selection under certain circumstances is desired. Moreover, the registration cost of HMIPv6 needs to be reduced further. However, the performance of existing mobility based schemes is limited because they all neglected the traffic parameter of the mobile node (MN). In this paper, we propose an efficient three-level hierarchical architecture for HMIPv6 networks, in which an MN may register with either a higher/lower Mobility Anchor Point (MAP) or its home agent. By analyzing the inter-domain registration cost function, we obtain a novel parameter for MAP selection, which takes both the mobility parameter and the traffic parameter of an MN into account to minimize the overall registration cost. Simulation results show that normally the novel MAP selection algorithm achieves 25% percentage reduction of registration cost over the traditional mobility based algorithm. In addition, this algorithm obtains better performance when the mobility parameter and the traffic parameter of the MN vary.