Key Technologies in Mobile Visual Search and MPEG Standardization Activities

Visual search has been a long-standing problem in applications such as location recognition and product search. Much research has been done on image representation, matching, indexing, and retrieval. Key component technologies for visual search have been developed, and numerous real-world applications are emerging. To ensure application interoperability, the Moving Picture Experts Group （MPEG） has begun standardizing visuaJ search technologies and is developing the compact descriptors for visua） search （CDVS） standard. MPEG seeks to develop a collaborative platform for evaluating existing visual search technologies. Peking University has participated in this standardization since the 94th MPEG meeting, and significant progress has been made with the various proposals. A test model （TM） has been selected to determine the basic pipeline and key components of visual search. However, the first-version TM has high computational complexity and imperfect retrieval and matching. Core experiments have therefore been set up to improve TM. In this article, we summarize key technologies for visual search and report the progress of MPEG CDVS. We discuss Peking University＇ s efforts in CDVS and also discuss unresolved issues.

AVS 3D Video Coding Technology and System

Following the success of the audio video standard （AVS） for 2D video coding, in 2008, the China AVS workgroup started developing 3D video （3DV） coding techniques. In this paper, we discuss the background, technical features, and applications of AVS 3DV coding technology. We introduce two core techniques used in AVS 3DV coding： inter-view prediction and enhanced stereo packing coding. We elaborate on these techniques, which are used in the AVS real-time 3DV encoder. An application of the AVS 3DV coding system is presented to show the great practical value of this system. Simulation results show that the advanced techniques used in AVS 3DV coding provide remarkable coding gain compared with techniques used in a simulcast scheme.

A ranking SVM based fusion model for cross-media meta-search engine

Recently,we designed a new experimental system MSearch,which is a cross-media meta-search system built on the database of the WikipediaMM task of ImageCLEF 2008.For a meta-search engine,the kernel problem is how to merge the results from multiple member search engines and provide a more effective rank list.This paper deals with a novel fusion model employing supervised learning.Our fusion model employs ranking SVM in training the fusion weight for each member search engine. We assume the fusion weight of each member search engine as a feature of a result document returned by the meta-search engine. For a returned result document,we first build a feature vector to represent the document,and set the value of each feature as the document's score returned by the corresponding member search engine.Then we construct a training set from the documents returned from the meta-search engine to learn the fusion parameter.Finally,we use the linear fusion model based on the overlap set to merge the results set.Experimental results show that our approach significantly improves the performance of the cross-media meta-search(MSearch) and outperforms many of the existing fusion methods.

(2+1)-Dimensional Spatial Localized Modes in Cubic-Quintic Nonlinear Media with the PT-Symmetric Potentials

We obtain exact spatial localized mode solutions of a(2+1)-dimensional nonlinear Schr¨odinger equation with constant diffraction and cubic-quintic nonlinearity in PT-symmetric potential, and study the linear stability of these solutions. Based on these results, we further derive exact spatial localized mode solutions in a cubic-quintic medium with harmonic and PT-symmetric potentials. Moreover, the dynamical behaviors of spatial localized modes in the exponential diffraction decreasing waveguide and the periodic distributed amplification system are investigated.

Wearable sensors for 3D upper limb motion modeling and ubiquitous estimation

Human motion capture technologies are widely used in interactive game and learning, animation, film special effects, health care, and navigation. Because of the agility, upper limb motion estimation is the most difficult problem in human motion capture. Traditional methods always assume that the movements of upper arm and forearm are independent and then estimate their movements separately; therefore, the estimated motion are always with serious distortion. In this paper, we propose a novel ubiquitous upper limb motion estimation method using wearable microsensors, which concentrates on modeling the relationship of the movements between upper arm and forearm. Exploration of the skeleton structure as a link structure with 5 degrees of freedom is firstly proposed to model human upper limb motion. After that, parameters are defined according to Denavit-Hartenberg convention, forward kinematic equations of upper limb are derived, and an unscented Kalman filter is invoked to estimate the defined parameters. The experimental results have shown the feasibility and effectiveness of the proposed upper limb motion capture and analysis algorithm.