Dispersion Constraint Based Non-negative Sparse Coding Model

Based on advantages of basic non-negative sparse coding (NNSC) model, and considered the prior class constraint of image features, a novel NNSC model is discussed here. In this NNSC model, the sparseness criteria is selected as a two-parameter density estimation model and the dispersion ratio of within-class and between-class is used as the class constraint. Utilizing this NNSC model, image features can be extracted successfully. Further, the feature recognition task by using different classifiers can be implemented well. Simulation results prove that our NNSC model proposed is indeed effective in extracting image features and recognition task in application.     

Strong Convergence Analysis of Batch Gradient-Based Learning Algorithm for Training Pi-Sigma Network Based on TSK Fuzzy Models

By combining of the benefits of high-order network and TSK (Tagaki-Sugeno-Kang) inference system, Pi-Sigma network is capable to dispose with the nonlinear problems much more effectively, which means it has a compacter construction, and quicker computational speed. The aim of this paper is to present a gradient-based learning method for Pi-Sigma network to train TSK fuzzy inference system. Moreover, some strong convergence results are established based on the weak convergence outcomes, which indicates that the sequence of weighted fuzzy parameters gets to a fixed point. Simulation results show the modified learning algorithm is effective to support the theoretical results.     

An SNN-Based Semantic Role Labeling Model with Its Network Parameters Optimized Using an Improved PSO Algorithm

Semantic role labeling (SRL) is a fundamental task in natural language processing to find a sentence-level semantic representation. The semantic role labeling procedure can be viewed as a process of competition between many order parameters, in which the strongest order parameter will win by competition and the desired pattern will be recognized. To realize the above-mentioned integrative SRL, we use synergetic neural network (SNN). Since the network parameters of SNN directly influence the synergetic recognition performance, it is important to optimize the parameters. In this paper, we propose an improved particle swarm optimization (PSO) algorithm based on log-linear model and use it to effectively determine the network parameters. Our contributions are two-folds: firstly, a log-linear model is introduced to PSO algorithm which can effectively make use of the advantages of a variety of different knowledge sources, and enhance the decision making ability of the model. Secondly, we propose an improved SNN model based on the improved PSO and show its effectiveness in the SRL task. The experimental results show that the proposed model has a higher performance for semantic role labeling with more powerful global exploration ability and faster convergence speed, and indicate that the proposed model has a promising future for other natural language processing tasks.     

Neural Networks Based PID Control of Bidirectional Inductive Power Transfer System

Inductive power transfer (IPT) systems facilitate contactless power transfer between two sides and across an air-gap, through weak magnetic coupling. However, IPT systems constitute a high order resonant circuit and, as such, are difficult to design and control. Aiming at the control problems for bidirectional IPT system, a neural networks based proportional-integral-derivative (PID) control strategy is proposed in this paper. In the proposed neural PID method, the PID gains, \(K_{P}\), \(K_{I}\) and \(K_{D}\) are treated as Gaussian potential function networks (GPFN) weights and they are adjusted using online learning algorithm. In this manner, the neural PID controller has more flexibility and capability than conventional PID controller with fixed gains. The convergence of the GPFN weights learning is guaranteed using Lyapunov method. Simulations are used to test the effective performance of the proposed controller.     

3D Surface Reconstruction of Noisy Point Clouds Using Growing Neural Gas: 3D Object/Scene Reconstruction

With the advent of low-cost 3D sensors and 3D printers, scene and object 3D surface reconstruction has become an important research topic in the last years. In this work, we propose an automatic (unsupervised) method for 3D surface reconstruction from raw unorganized point clouds acquired using low-cost 3D sensors. We have modified the growing neural gas network, which is a suitable model because of its flexibility, rapid adaptation and excellent quality of representation, to perform 3D surface reconstruction of different real-world objects and scenes. Some improvements have been made on the original algorithm considering colour and surface normal information of input data during the learning stage and creating complete triangular meshes instead of basic wire-frame representations. The proposed method is able to successfully create 3D faces online, whereas existing 3D reconstruction methods based on self-organizing maps required post-processing steps to close gaps and holes produced during the 3D reconstruction process. A set of quantitative and qualitative experiments were carried out to validate the proposed method. The method has been implemented and tested on real data, and has been found to be effective at reconstructing noisy point clouds obtained using low-cost 3D sensors.     

DCT statistics-based digital dropout detection in degraded archived media

With the rapid development of visual digital media, the demand for better quality of service has increased the pressure on broadcasters to automate their error detection and restoration activities for preserving their archives. Digital dropout is one of the defects that affect archived visual materials and tends to occur in block by block basis (size of 8 × 8). It is well established that human visual system (HVS) is highly adapted to the statistics of its visual natural environment. Consequently, in this paper, we have formulated digital dropout detection as a classification problem which predicts block label based on statistical features. These statistical features are indicative of perceptual quality relevant to human visual perception, and allow pristine images to be distinguished from distorted ones. Here, the idea is to extract discriminant block statistical features based on discrete cosine transform (DCT) coefficients and determine an optimal neighborhood sampling strategy to enhance the discrimination ability of block representation. Since this spatial frame based approach is free from any motion computation dependency, it works perfectly in the presence of fast moving objects. Experiments are performed on video archives to evaluate the efficacy of the proposed technique.     

PCET based copy-move forgery detection in images under geometric transforms

With the advent of the powerful editing software and sophisticated digital cameras, it is now possible to manipulate images. Copy-move is one of the most common methods for image manipulation. Several methods have been proposed to detect and locate the tampered regions, while many methods failed when the copied region undergone some geometric transformations before being pasted, because of the de-synchronization in the searching procedure. This paper presents an efficient technique for detecting the copy-move forgery under geometric transforms. Firstly, the forged image is divided into overlapping circular blocks, and Polar Complex Exponential Transform (PCET) is employed to each block to extract the invariant features, thus, the PCET kernels represent each block. Secondly, the Approximate Nearest Neighbor (ANN) Searching Problem is used for identifying the potential similar blocks by means of locality sensitive hashing (LSH). In order to make the algorithm more robust, morphological operations are applied to remove the wrong similar blocks. Experimental results show that our proposed technique is robust to geometric transformations with low computational complexity.     

Landmark-based music recognition system optimisation using genetic algorithms

Audio fingerprinting allows us to label an unidentified music fragment within a previously generated database. The use of spectral landmarks aims to obtain a robustness that lets a certain level of noise be present in the audio query. This group of audio identification algorithms holds several configuration parameters whose values are usually chosen based upon the researcher’s knowledge, previous published experimentation or just trial and error methods. In this paper we describe the whole optimisation process of a Landmark-based Music Recognition System using genetic algorithms. We define the actual structure of the algorithm as a chromosome by transforming its high relevant parameters into various genes and building up an appropriate fitness evaluation method. The optimised output parameters are used to set up a complete system that is compared with a non-optimised one by designing an unbiased evaluation model.