Recent real-time semantic segmentation models, whether single-branch or multi-branch, achieve good performance and speed. However, their speed is limited by multi-path blocks, and some depend on high-performance teacher models for training. To overcome these issues, we propose Golden Cudgel Network (GCNet). Specifically, GC-Net uses vertical multi-convolutions and horizontal multi-paths for training, which are reparameterized into a single convolution for inference, optimizing both performance and speed. This design allows GCNet to self-enlarge during training and self-contract during inference, effectively becoming a "teacher model" without needing external ones. Experimental results show that GCNet outperforms existing state-of-the-art models in terms of performance and speed on the Cityscapes, CamVid, and Pascal VOC 2012 datasets. The code is available at https://github.com/gyyang23/GCNet.

Real-world low-light images suffer from two main degradations, namely, inevitable noise and poor visibility. Since the noise exhibits different levels, its estimation has been implemented in recent works when enhancing low-light images from raw Bayer space. When it comes to sRGB color space, the noise estimation becomes more complicated due to the effect of the image processing pipeline. Nevertheless, most existing enhancing algorithms in sRGB space only focus on the low visibility problem or suppress the noise under a hypothetical noise level, leading them impractical due to the lack of robustness. To address this issue, we propose an adaptive unfolding total variation network (UTVNet), which approximates the noise level from the real sRGB low-light image by learning the balancing parameter in the model-based denoising method with total variation regularization. Meanwhile, we learn the noise level map by unrolling the corresponding minimization process for providing the inferences of smoothness and fidelity constraints. Guided by the noise level map, our UTVNet can recover finer details and is more capable to suppress noise in real captured low-light scenes. Extensive experiments on real-world low-light images clearly demonstrate the superior performance of UTVNet over state-of-the-art methods.

Image enhancement aims to improve the aesthetic quality of images. Most enhancement methods are based on image decomposition techniques. For example, an entire image can be decomposed into a smooth base layer and a residual detail layer. Applying appropriate algorithms to different layers can solve most enhancement problems. Besides decomposing the entire image, the local decomposition approach in local Laplacian filter can also achieve satisfied enhancement results. As a standard convolution is also a local operator that the output values is determined by neighborhood pixels, we observe that the standard convolution can be improved by integrating the local decomposition method for better solving image enhancement problems. Based on this analysis, we propose Windowing Decomposition Convolution (WDC) that decomposes the content of each convolution window by a windowing basic value before applying convolution operation. Using different windowing basic values, the WDC can gather global information and locally separate the processing of different components of images. Moreover, combined with WDC, a new Windowing Decomposition Convolutional Neural Network (WDCNN) is presented. The experimental results show that our WDCNN achieves superior enhancement performance on the MIT-Adobe FiveK and sRGB-SID datasets for noise-free image retouching and low-light noisy image enhancement compared with state-of-the-art techniques.

In this article, a novel fuzzy-based dynamic multiobjective evolutionary algorithm is presented. In this article, giving a valid and true response to the change is an essential task to improve the diversity of solutions when an environmental change occurs. The basic idea is to randomly remove some solutions and replace by newly created solutions. However, the random selection detours the algorithm's trajectory and deteriorates the performance of the optimization algorithm. Recently, the Borda method has been deployed to find the best candidates to be removed from the solutions list. Although the Borda method outperforms the random strategy, it suffers from some drawbacks. In this article, we propose an improved Borda count method incorporated with fuzzy tuned parameters so that its parameters are adjusted by Mamdani fuzzy rules. Our new Borda method can distinguish the information before and after change with different fuzzy weights. In addition to the fuzzy-based Borda, we employ an improved evolutionary algorithm based on fuzzy logic. We propose a novel nondominated sorting genetic algorithm with its parameters tuned with fuzzy rules so that it is adapted to the new environment. Experiments are conducted on standard benchmarks and the results are compared with recent algorithms. Then, multimedia data analysis, such as segmentation of moving objects, is experimented as a dynamic multiobjective problem and solved by the proposed algorithm.

Learning from a limited number of labeled samples (pixels) remains a key challenge in the hyperspectral image (HSI) classification. To address this issue, we propose a deep metric learning-based feature embedding model, which can meet the tasks both for same- and cross-scene HSI classifications. In the first task, when only a few labeled samples are available, we employ ideas from metric learning based on deep embedding features and make a similarity learning between pairs of samples. In this case, the proposed model can learn well to compare whether two samples belong to the same class. In another task, when an HSI image (target scene) that needs to be classified is not labeled at all, the embedding model can learn from another similar HSI image (source scene) with sufficient labeled samples and then transfer to the target model by using an unsupervised domain adaptation technique, which not only employs the adversarial approach to make the embedding features from the source and target samples indistinguishable but also encourages the target scene's embeddings to form similar clusters with the source scene one. After the domain adaptation between the HSIs of the two scenes is finished, any traditional HSI classifier can be used. In a simple manner, the nearest neighbor (NN) algorithm is selected as the classifier for the classification tasks throughout this article. The experimental results from a series of popular HSIs demonstrate the advantages of the proposed model both in the same- and cross-scene classification tasks.

In this paper, we present a simple and effective framework called Occlusion-adaptive Deep Networks (ODN) with the purpose of solving the occlusion problem for facial landmark detection. In this model, the occlusion probability of each position in high-level features are inferred by a distillation module that can be learnt automatically in the process of estimating the relationship between facial appearance and facial shape. The occlusion probability serves as the adaptive weight on high-level features to reduce the impact of occlusion and obtain clean feature representation. Nevertheless, the clean feature representation cannot represent the holistic face due to the missing semantic features. To obtain exhaustive and complete feature representation, it is vital that we leverage a low-rank learning module to recover lost features. Considering that facial geometric characteristics are conducive to the low-rank module to recover lost features, we propose a geometry-aware module to excavate geometric relationships between different facial components. Depending on the synergistic effect of three modules, the proposed network achieves better performance in comparison to state-of-the-art methods on challenging benchmark datasets.

Facial landmark detection has witnessed substantial progress due to introducing convolutional neural networks. Nonetheless, current convolutional neural networks-based approaches ignore the useful geometric relationship between different facial locations. To address this issue, we propose a new module to model the facial geometric relationship. The module can be integrated into the convolutional neural networks architecture to obtain the geometric representation, whereafter we leverage bilinear pooling operation to embed it into high-level feature maps of original face image so as to produce the more discriminative face representation. Extensive evaluation experiments on multiple challenging benchmark datasets demonstrate that our captured geometric information is robust against occlusion and head pose variation and our proposed method outperforms state-of-the-art methods.

In this paper, we propose a new and effective multi-objective optimization algorithm based on a modified harmony search. The proposed method employs reverse learning in the harmony vector updating equation in order to enhance the global searching ability. Moreover, it adopts a harmony anchoring scheme so that unnecessary exploration is avoided. Experimental studies carried on eight benchmark problems show quite satisfactory results and indicate the higher performance of the proposed algorithm in comparison with traditional multi-objective optimization algorithms. Finally, it has been applied to solve the image segmentation problem.

This paper introduces a novel dynamic multi-objective optimization algorithm. When a change occurs, the optimization algorithm has to give a true response to change. One of the most popular approaches of responding to change is to select some solutions randomly and remove them from the population. Since random selection results in deterioration of the algorithm's performance, the wise selection is crucial. Recently, the Borda count method was applied to find the most appropriate solutions to be eliminated from the population. However, basic Borda suffers from the same weight of information before and after the change. In this paper, we propose a new weighted Borda count so that its parameters are tuned by fuzzy rules. Mamdani fuzzy rules have been employed to tune the weights and distinguish between information before and after the change. Finally, the 'Change Effect' is proposed to calculate the effect of the change on the solutions. The performance of the proposed algorithm is tested on standard functions and is compared with recent algorithms.

Low-Rank Representation (LRR) is one of the most effective algorithms for subspace clustering. Exploring the multiple subspace structures of data such as Low-Rank Representation is effective in subspace clustering. Low-Rank Representation is usually formulated as a constrained convex optimization problem regularized with nuclear norm. It needs singular values decomposition (SVD) in every iteration which is challenging in terms of time complexity and memory. Recently, its non-convex formulation, NLRR [1] via matrix factorization framework has been proposed, and becomes one of state-of-the-art techniques for subspace clustering. However, NLRR cannot scale to problems with large n (number of samples) since it requires either the inversion of an n × n matrix or solving an n × n linear system. In this paper, we develop a faster algorithm for solving NLRR problem, with time complexity per iteration reduced from (n3) to (mnd) and memory complexity from (n2) to (mn), where m is dimensionality and d is the target rank (usually d ≪ m ≪ n). The main idea of our algorithm is to reformulate NLRR as a sum of rank-one components and apply a column-wise block coordinate descent to update each component iteratively. Our new method is considerably faster and more accurate in practice for subspace clustering. We also show that the proposed method is guaranteed to converge to stationary points. Moreover, considering the high demand in memory and computational time for the final spectral clustering phase, we also propose an efficient clustering approach which can further boost the performance of subspace clustering. Experiments on extensive simulations and real datasets confirm the efficiency of our proposed NLRR++. In particular, we are more than 12 times faster than state-of-the-art subspace clustering method, NLRR, and our method is the only subspace clustering algorithm that can scale to the Imagenet dataset with 120K samples.

Abstract The main goal of this paper is to introduce a novel dynamic multi‐objective optimization algorithm. First, after detecting the environmental changes, Borda count ranking method is applied to population in order to assign the Borda score to each individual, and then the lowest score individuals are removed from population and replaced with new created solutions. Furthermore, fuzzy adaptive multi‐objective cat swarm optimization algorithm is used to estimate the Pareto‐optimal front in which its parameters are tuned to new environment by Mamdani fuzzy rules when a change occurs. Performance of the proposed algorithm is tested on dynamic multi‐objective benchmarks and is compared with recent achievements. The simulations show the quite satisfactory results and higher performance of the proposed method in comparison with traditional approaches.

Learning robust regression model from high-dimensional corrupted data is an essential and difficult problem in many practical applications. The state-of-the-art methods have studied low-rank regression models that are robust against typical noises (like Gaussian noise and out-sample sparse noise) or outliers, such that a regression model can be learned from clean data lying on underlying subspaces. However, few of the existing low-rank regression methods can handle the outliers/noise lying on the sparsely corrupted disjoint subspaces. To address this issue, we propose a low-rank-sparse subspace representation for robust regression, hereafter referred to as LRS-RR in this paper. The main contribution include the following: (1) Unlike most of the existing regression methods, we propose an approach with two phases of low-rank-sparse subspace recovery and regression optimization being carried out simultaneously,(2) we also apply the linearized alternating direction method with adaptive penalty to solved the formulated LRS-RR problem and prove the convergence of the algorithm and analyze its complexity, (3) we demonstrate the efficiency of our method for the high-dimensional corrupted data on both synthetic data and two benchmark datasets against several state-of-the-art robust methods.

We present a novel approach for automatically detecting and tracking facial landmarks across poses and expressions from in-the-wild monocular video data, e.g., YouTube videos and smartphone recordings. Our method does not require any calibration or manual adjustment for new individual input videos or actors. Firstly, we propose a method of robust 2D facial landmark detection across poses, by combining shape-face canonical-correlation analysis with a global supervised descent method. Since 2D regression-based methods are sensitive to unstable initialization, and the temporal and spatial coherence of videos is ignored, we utilize a coarse-todense 3D facial expression reconstruction method to refine the 2D landmarks. On one side, we employ an in-the-wild method to extract the coarse reconstruction result and its corresponding texture using the detected sparse facial landmarks, followed by robust pose, expression, and identity estimation. On the other side, to obtain dense reconstruction results, we give a face tracking flow method that corrects coarse reconstruction results and tracks weakly textured areas; this is used to iteratively update the coarse face model. Finally, a dense reconstruction result is estimated after it converges. Extensive experiments on a variety of video sequences recorded by ourselves or downloaded from YouTube show the results of facial landmark detection and tracking under various lighting conditions, for various head poses and facial expressions. The overall performance and a comparison with state-of-art methods demonstrate the robustness and effectiveness of our method.

Abstract Face alignment and tracking play important roles in facial performance capture. Existing data‐driven methods for monocular videos suffer from large variations of pose and expression. In this paper, we propose an efficient and robust method for this task by introducing a novel supervised coordinate descent method with 3D bilinear representation. Instead of learning the mapping between the whole parameters and image features directly with a cascaded regression framework in current methods, we learn individual sets of parameters mappings separately step by step by a coordinate descent mean. Because different parameters make different contributions to the displacement of facial landmarks, our method is more discriminative to current whole‐parameter cascaded regression methods. Benefiting from a 3D bilinear model learned from public databases, the proposed method can handle the head pose changes and extreme expressions out of plane better than other 2D‐based methods. We present the reliable result of face tracking under various head poses and facial expressions on challenging video sequences collected online. The experimental results show that our method outperforms state‐of‐art data‐driven methods.

Person re-identification, aiming to identify images of the same person from various cameras views in different places, has attracted a lot of research interests in the field of artificial intelligence and multimedia. As one of its popular research directions, the metric learning method plays an important role for seeking a proper metric space to generate accurate feature comparison. However, the existing metric learning methods mainly aim to learn an optimal distance metric function through a single metric, making them difficult to consider multiple similar relationships between the samples. To solve this problem, this paper proposes a coarse-to-fine deep metric learning method equipped with multiple different Stacked Auto-Encoder (SAE) networks and classification networks. In the perspective of the human's visual mechanism, the multiple different levels of deep neural networks simulate the information processing of the brain's visual system, which employs different patterns to recognize the character of objects. In addition, a weighted assignment mechanism is presented to handle the different measure manners for final recognition accuracy. The experimental results conducted on two public datasets, i.e., VIPeR and CUHK have shown the prospective performance of the proposed method.

Dimensionality reduction (DR) has been considered as one of the most significant tools for data analysis. One type of DR algorithms is based on latent variable models (LVM). LVM-based models can handle the preimage problem easily. In this paper we propose a new LVM-based DR model, named thin plate spline latent variable model (TPSLVM). Compared to the well-known Gaussian process latent variable model (GPLVM), our proposed TPSLVM is more powerful especially when the dimensionality of the latent space is low. Also, TPSLVM is robust to shift and rotation. This paper investigates two extensions of TPSLVM, i.e., the back-constrained TPSLVM (BC-TPSLVM) and TPSLVM with dynamics (TPSLVM-DM) as well as their combination BC-TPSLVM-DM. Experimental results show that TPSLVM and its extensions provide better data visualization and more efficient dimensionality reduction compared to PCA, GPLVM, ISOMAP, etc.

A novel application of the Hough transform (HT) neighborhood approach to collinear segment detection was proposed in [1]. It, however, suffered from one major weakness in that it could not provide an effective solution to the case of segment intersection. This paper analyzes a vital prerequisite step, disturbance elimination in the Hough space, and shows why, this method alone, is incapable of distinguishing the true segment endpoints. To address the problem, a unique HT butterfly separation method is proposed in this correspondence, as an essential complement to the above publication.

In this paper, we extend our previously proposed line detection method to line segmentation using a so-called unite-and-divide (UND) approach. The methodology includes two phases, namely the union of spectra in the frequency domain, and the division of the sinogram in Radon space. In the union phase, given an image, its sinogram is obtained by parallel 2D multilayer Fourier transforms, Cartesian-to-polar mapping and 1D inverse Fourier transform. In the division phase, the edges of butterfly wings in the neighborhood of every sinogram peak are firstly specified, with each neighborhood area corresponding to a window in image space. By applying the separated sinogram of each such windowed image, we can extract the line segments. The division Phase identifies the edges of butterfly wings in the neighborhood of every sinogram peak such that each neighborhood area corresponds to a window in image space. Line segments are extracted by applying the separated sinogram of each windowed image. Our experiments are conducted on benchmark images and the results reveal that the UND method yields higher accuracy, has lower computational cost and is more robust to noise, compared to existing state-of-the-art methods.

Dimensionality reduction (DR) has been considered as one of the most significant tools for data analysis. In this paper we propose a new latent variable model based on the thin plate splines, named Thin Plate Spline Latent Variable Model (TPSLVM). It has strong connection with the so-called Gaussian Process Latent Variable Model (GPLVM). We demonstrate that the proposed TPSLVM can be viewed as the GPLVM with a fairly peculiar covariance function. Moreover, compared to GPLVM, TPSLVM is more powerful especially when the dimensionality of the latent space is very low (e.g., 2D or 3D). One of main purposes of DR algorithms is to visualize data in 2D/3D spaces. Therefore, TPSLVM will benefit this process. Experimental results show that TPSLVM provides better data visualization and more efficient dimensionality reduction than GPLVM.

A maximum-amplitude-based coarse-to-fine algorithm is proposed with two novel ideas highlighted: adopting the maximum value of the fractional Fourier amplitude spectrum to measure the compactness of a signal, and using a coarse-to-fine strategy to speed up the searching process. The simulation results on synthetic and real signals show the validity of the proposed method.

The Hough transform (HT) is a commonly used technique for the identification of straight lines in an image. The Hough transform can be equivalently computed using the Radon transform (RT), by performing line detection in the frequency domain through use of central-slice theorem. In this research, an advanced Radon transform is developed using a multilayer fractional Fourier transform, a Cartesian-to-polar mapping, and 1-D inverse Fourier transforms, followed by peak detection in the sinogram. The multilayer fractional Fourier transform achieves a more accurate sampling in the frequency domain, and requires no zero padding at the stage of Cartesian-to-polar coordinate mapping. Our experiments were conducted on mix-shape images, noisy images, mixed-thickness lines and a large data set consisting of 751,000 handwritten Chinese characters. The experimental results have shown that our proposed method outperforms all known representative line detection methods based on the standard Hough transform or the Fourier transform.

Inspired by the philosophy of ancient Chinese Taoism, Xu's Bayesian ying-yang (BYY) learning technique performs clustering by harmonizing the training data (yang) with the solution (ying). In our previous work, the BYY learning technique was applied to a fuzzy cerebellar model articulation controller (FCMAC) to find the optimal fuzzy sets; however, this is not suitable for time series data analysis. To address this problem, we propose an incremental BYY learning technique in this paper, with the idea of sliding window and rule structure dynamic algorithms. Three contributions are made as a result of this research. First, an online expectation-maximization algorithm incorporated with the sliding window is proposed for the fuzzification phase. Second, the memory requirement is greatly reduced since the entire data set no longer needs to be obtained during the prediction process. Third, the rule structure dynamic algorithm with dynamically initializing, recruiting, and pruning rules relieves the "curse of dimensionality" problem that is inherent in the FCMAC. Because of these features, the experimental results of the benchmark data sets of currency exchange rates and Mackey-Glass show that the proposed model is more suitable for real-time streaming data analysis.

Search Engine has proven its effectiveness for retrieval of information from World Wide Web. Traditionally, the search results are arranged in an ordered list by popularity and relevancy. However, the enormous size of matched Web pages causes inefficiency for users to locate the most relevant Web pages. A proper organization of the search result is important to improve its browsability of Web searching. In this paper, we proposed by performing Support Vector Clustering (SVC) on the search result to reorganize results in groups of similar context to facilitate effective browsing of search result by the users. SVC is a nonparametric clustering algorithm that can group clusters with arbitrary shapes and without the need to specify the number of clusters. It is a kernel clustering method that maps via a nonlinear function to a high dimension feature space. To obtain the optimal clustering result, choosing of the accurate parameters (kernel width and penalty coefficient) for SVC is crucial. In this paper, it proposed an automatic tuning method for SVC parameters to obtain the optimal result. The results from the experiment have proven the effectiveness and usefulness of above mentioned method. The performance is comparable to other popular clustering techniques.

A simple search keyword usually returns million of search results. The result count may appear impressive, at the same time it confuse the users. User usually will not wish to browse through million of entries. This paper proposed a query refinement method by iterative clustering of information from the web page content. QRSE system is developed to demonstrate the abovementioned concept.

This paper analyses stock market price prediction based on a hierarchical cerebellar model arithmetic controller (HCMAC) neural network. Applications using stock market price prediction tools are required to be adaptive to new incoming data as well as have fast learning capabilities. Current popular neural networks are based on the Multi Layer Perceptron (MLP) structure which has low memory consumption and has a fast processing speed however the performance of the MLP deteriorates as the network expands. An HCMAC structure uses a direct memory mapping technique which would perform consistently fast independent of size of network. The drawback is a huge amount of memory is required to perform direct mapping. This can be reduced by using self-organizing techniques to optimize the clusters during each training cycle. The accuracy of the output can be controlled based on the values set in the cyclic self-organizing module. The cyclic self-organizing HCMAC (CSOHCMAC) combines both the HCMAC and cyclic self-organizing modules to create a neural network model that would be robust and fast as well as flexible to adapt to changes

Inspired by the movement of comet under gravitational interaction with planets and stars, we propose a novel wavelet-gravitation network (WGN) for endoscope optical tracking. We substitute the world gravitation law with interaction described by radial wavelet function. The WGN is trained based on the energy conservation law. The prediction of target image position is made with movement described by the second Newton's law. The WGN is incorporated with Kalman filtering framework. Our experiments show that WGN performs especially well for nonlinear movement tracking. As the movement of surgical tools is hand-driven, it is mostly nonlinear and affected by both tangential and normal accelerations. Our proposed methodology has been successfully applied to endoscope tracking for image guided surgery

This paper proposes an evolving fuzzy associative memory neural network model based on the fuzzy CMAC (FCMAC). FCMAC is an auto-associate memory feed forward neural network with attractive properties of fast learning and simple computation. Evolving techniques aim at building adaptive intelligent systems that evolve both their structure and parameters through incremental online learning. During fuzzification phase, the proposed ESOFCMAC uses raw numerical values of a training data set without any preprocessing and obtains dynamic partition-base clusters with no prior knowledge of the number of clusters. The performance of ESOFCMAC is illustrated on several benchmark data sets and compared with traditional models.

As an associative memory neural network model, the cerebellar model articulation controller (CMAC) has attractive properties of fast learning and simple computation, but its rigid structure makes it difficult to approximate certain functions. This research attempts to construct a novel neural fuzzy CMAC, in which Bayesian Ying-Yang (BYY) learning is introduced to determine the optimal fuzzy sets, and a truth-value restriction inference scheme is subsequently employed to derive the truth values of the rule weights of implication rules. The BYY is motivated from the famous Chinese ancient Ying-Yang philosophy: everything in the universe can be viewed as a product of a constant conflict between opposites-Ying and Yang, a perfect status is reached when Ying and Yang achieve harmony. The proposed fuzzy CMAC (FCMAC)-BYY enjoys the following advantages. First, it has a higher generalization ability because the fuzzy rule sets are systematically optimized by BYY; second, it reduces the memory requirement of the network by a significant degree as compared to the original CMAC; and third, it provides an intuitive fuzzy logic reasoning and has clear semantic meanings. The experimental results on some benchmark datasets show that the proposed FCMAC-BYY outperforms the existing representative techniques in the research literature.

A real-time synthetic talking head provides a natural communication interface for human-machine interaction. In this paper, two driving schemes for Mandarin are proposed. One is a speech-driven method using wavelet network to map audio signals to visual movements, the other is a hybrid driving scheme making use of linguistic analysis. Based on this framework, a Web browser is successfully built up, in which a virtual newsreader can read Chinese Web articles in real time

This paper presents a post-processing system for improving the recognition rate of a Handwritten Chinese Character Recognition (HCCR) device. This three-stage hybrid post-processing system reduces the misclassification and rejection rates common in the single character recognition phase. The proposed system is novel in two respects: first, it reduces the misclassification rate by applying a dictionary-look-up strategy that bind the candidate characters into a word-lattice and appends the linguistic-prone characters into the candidate set; second, it identifies promising sentences by employing a distant Chinese word BI-Gram model with a maximum distance of three to select plausible words from the word-lattice. These sentences are then output as the upgraded result. Compared with one of our previous works in single Chinese character recognition, the proposed system improves absolute recognition rates by 12%.

In this paper, we present an evolutionary framework for efficient aerodynamic shape design. The approach suggests employing hybrid evolutionary algorithm with gradient-based local search method in the spirit of Lamarckian and surrogate models that approximates the computationally expensive adjoint computational fluid dynamics during design search. In particular, we reveal that the proposed framework guarantees global convergence by inheriting the properties of trust-region method to interleave use of the exact solver for the objective function with computationally cheap surrogate models during local search. Empirical results on 2D airfoil shape design using an adjoint inverse pressure design problem indicates that the approaches global convergences on a limited computational budget.

This paper proposes active handwriting models, in which kernel principal component analysis is applied to capture nonlinear handwriting variations. In the recognition phase, the chamfer distance transform and a dynamic tunneling algorithm (DTA) are employed to search for the optimal shape parameters. The proposed methodology is successfully applied to a novel radical decomposition approach to the challenging problem of handwritten Chinese character recognition.

This paper describes a kernel active handwriting model (K-AHM) and its application to handwritten Chinese character recognition. In the model, the kernel principal component analysis is applied to capture nonlinear variations caused by handwriting, and a fitness function on the basis of a chamfer distance transform is introduced to search for optimal shape parameters using genetic algorithms (GAs). The K-AHM is applied to handwritten Chinese character recognition, which converts the complex pattern recognition problem into recognizing a small set of primitive structures called radicals. By treating Chinese character composition as a discrete-time Markov process, character composition is carried out with the Viterbi algorithm. The proposed methodology has been successfully implemented in an experimental recognition system.

In this paper, entropy is a term used in the learning phase of a neural network. As learning progresses, more hidden nodes get into saturation. The early creation of such hidden nodes may impair generalisation. Hence an entropy approach is proposed to dampen the early creation of such nodes by using a new computation called entropy cycle. Entropy learning also helps to increase the importance of relevant nodes while dampening the less important nodes. At the end of learning, the less important nodes can then be pruned to reduce the memory requirements of the neural network.

Handwritten Chinese characters can be recognized by first extracting the basic shapes (radicals) of which they are composed. Radicals are described by nonlinear active shape models and optimal parameters found using the chamfer distance transform and a dynamic tunneling algorithm. The radical recognition rate is 96.5 percent correct (writer-independent) on 280,000 characters containing 98 radical classes.

Offline handwritten Chinese character recognition is a very hard pattern-recognition problem of considerable practical importance. Two popular approaches are to extract features holistically from the character image or to decompose characters structurally into component parts---usually strokes. Here we take a novel approach, that of decomposing into radicals on the basis of image information (i.e., without first decomposing into strokes). During training, 60 examples of each radical were represented by "landmark" points, labeled semiautomatically, with radicals in different characteristic positions treated as distinctly different radicals. Kernel principal-component analysis then captured the main (nonlinear) variations around the mean radical. During the recognition, the dynamic tunneling algorithm was used to search for optimal shape parameters in terms of chamfer distance minimization. Considering character composition as a Markov process in which up to four radicals are combined in some assumed sequential order, we can recognize complete, hierarchically-composed characters by using the Viterbi algorithm. This gave a character recognition rate of 93.5% characters correct (writer-independent) on a test set of 430,800 characters from 2,154 character classes composed of 200 radical categories, which is comparable to the best reported results in the literature. Although the initial semiautomatic landmark labeling is time consuming, the decomposition approach is theoretically well-motivated and allows the different sources of variability in Chinese handwriting to be handled separately and by the most appropriate means--either learned from example data or incorporated as prior knowledge. Hence, high generalizability is obtained from small amounts of training data, and only simple prior knowledge needs to be incorporated, thus promising robust recognition performance. As such, there is very considerable potential for further development and improvement in the direction of larger character sets and less constrained writing conditions.

This paper applies active handwriting models (AHM) to handwritten Chinese character recognition. Exploiting active shape models (ASM), the AHM can capture the handwriting variation from character skeletons. The AHM has the following characteristics: principal component analysis is applied to capture variations caused by handwriting, an energy functional on the basis of chamfer distance transform is introduced as a criterion to fit the model to a target character skeleton, and the dynamic tunneling algorithm (DTA) is incorporated with gradient descent to search for shape parameters. The AHM is used within a radical approach to handwritten Chinese characters recognition, which converts the complex pattern recognition problem to recognizing a small set of primitive structures-radicals. Our initial experiments are conducted on 98 radicals covering 1400 loosely-constrained Chinese character categories written by 200 different writers. The correct matching rate is 94.2% on these 2.8/spl times/10/sup 5/ characters. Comparison with existing radical approaches shows that our method achieves superior performance.

Handwritten Chinese character recognition is one of the most difficult problems of pattern recognition. Since the majority of Chinese characters are made up from just a small set of primitive structures (radicals), this paper describes an approach to active radical modeling for such handwritten characters. The most significant characteristic of our method is that radicals can be found robustly without stroke extraction, and the principal variations of the radical can be encoded in a small number of parameters. In the training phase, the example radicals are represented by manually-labeled 'landmark' points. Then a small number of principal components of the eigenvectors are calculated to capture the main variation of the training examples from the mean radical. In the matching phase, each radical model is fitted to the image evidence by adjusting the shape parameters in terms of chamfer distance minimization. Initial experiments are conducted on 1,100 loosely-constrained Chinese character categories written by 200 different writers. The correct matching rate is 95.8%, showing that our radical modeling is effective and capable of forming a sound basis for handwritten Chinese character recognition.

Supervised learning of Neocognitron is fulfilled by presenting training patterns, which map to specified features. However, the training patterns and many parameters are designed empirically and set manually in Fukushima's Neocognitron. In this paper, we use genetic algorithms (GAs) to tune the parameters of Neocognitron and search its reasonable training pattern sets. First of all, the correlation amongst the training patterns is considered as a critical factor affecting Neocognitron's performance, but it is ignored in the design of the original Neocognitron. Then, a GA-based supervised learning of the Neocognitron is proposed to tune the parameters and search training patterns. The results prove that the performance of a Neocognitron is sensitive to its training patterns, selectivity and receptive fields, and can be improved by this supervised learning on the basis of GAs and correlation analysis.

Optical character recognition (OCR) provides a solution to acquire, archive and retrieve a large amount of paper-based information which is still commonly used in our daily life. The process of a classical optical character recognition system consists of a series of stages, such as format analysis, text segmentation, feature extraction and classification. This paper focuses on the last two stages, and two contributions can be claimed: first, rapid transformed stroke density features (SDF) are used for preliminary classification and outline primitive structural features for final classification. Second, the original extension matrix algorithm is improved by heuristic path searching on the basis of information entropy as well as Laplace error rate evaluation function. Our experimental results prove that the rapid transformed SDFs are insensitive to image translation or rotation, and that the improved extension matrix algorithm outperforms other inductive approaches based on AE1 and AQ15. The excellent performance with respect to a large data set also indicates our proposed approach is effective and efficient.

In most models of mining fuzzy association rules, the items are considered to have equal importance. Due to diverse human interest and preference for items, such models do not work well in many situations. To improve such models, we propose a method to mine fuzzy association rules with weighted items. One of the major problems in data mining research is the development of good measures of interest of discovered rules. The weighted support and weighted confidence for fuzzy association rules are defined. Kohonen self-organized mapping is used to fuzzify the numerical attributes into linguistic terms. A new fuzzy association rule mining algorithm, which generalizes the popular Apriori Gen large itemset based algorithm, is developed. The advantages of the new algorithm are shown by testing it on a census database with 5000 transaction records.

The further study on the sensitivity analysis of Neocognitron is discussed in this paper. Fukushima's Neocognitron is capable of recognizing distorted patterns as well as tolerating positional shift. Supervised learning of the Neocognitron is fulfilled by training patterns layer by layer. However, many parameters, such as selectivity and receptive fields are set manually. Furthermore, in Fukushima's original Neocognitron, all the training patterns are designed empirically. In this paper, we use Genetic Algorithms (GAs) to tune the parameters of Neocognitron and search its reasonable training pattern sets. Four contributions are claimed: first, by analyzing the learning mechanism of Fukushima's original Neocognitron, the correlations amongst the training patterns are claimed to affect the performance of Neocognitron, tuning the Neocognitron's number of planes is equivalent to searching reasonable training patterns for its supervised learning; second, a GA-based supervised learning of the Neocognitron is carried out in this way, searching the parameters and training patterns by GAs but specifying the connection weights by training the Neocognitron; third, other than traditional GAs which are unsuitable for the large searching space of training patterns set, the cooperative coevolution is incorporated to play this role; fourth, an effective fitness function is given out when applying the above methodology into numeral recognition. The evolutionary computation in our initial experiments is implemented based on the original training pattern set, e.g. the individuals of the population are generated from Fukushima's original training patterns during initialization of GAs. The results prove that our correlation analysis is reasonable, and show that the performance of a Neocognitron is sensitive to its training patterns, selectivity and receptive fields, especially, the performance is not monotonically increasing with respect to the number of training patterns, and this GA-based supervised learning is able to improve Neocognitron's performance.

Feature selection is of great importance in recognition system design because it directly affects the overall performance of the recognition system. Feature selection can be considered as a problem of global combinatorial optimization. It is a very time-consuming task to search the most suitable features amongst a huge number of possible feature combinations, therefore, an effective and efficient search technique is desired. In this paper, we use genetic algorithms (GA) to design a feature selection approach for handwritten Chinese character recognition. Four contributions are claimed: First, the general transformed divergence among classes, which is derived from Mahalanobis distances, is proposed to be the fitness function in the feature selection based on GA; Second, a special crossover operator other than traditional one is given; Third, a special criterion of terminating selections is inferred from the criterion of minimum error probability in a Bayes classifier; Fourth, we compare our method with the feature selection based on branch-and-bound algorithm (BAB), which is often used to reduce the calculation of feature selection via exhaustive search. The analyses of the experimental results can be proceeded that traditional GA is an ergodic Markov chain, while, BAB is a depth first heuristic algorithm for exhaustive search. We conclude that the GA-based method proposed in this paper is promising to solve the feature selection problems in a multidimensional space.