Siu-Yeung Cho

A neural-based crowd estimation by hybrid global learning algorithm

A neural-based crowd estimation system for surveillance in complex scenes at underground station platform is presented. Estimation is carried out by extracting a set of significant features from sequences of images. Those feature indexes are modeled by a neural network to estimate the crowd density. The learning phase is based on our proposed hybrid of the least-squares and global search algorithms which are capable of providing the global search characteristic and fast convergence speed. Promising experimental results are obtained in terms of accuracy and real-time response capability to alert operators automatically.

Semantic colorization with internet images

Colorization of a grayscale photograph often requires considerable effort from the user, either by placing numerous color scribbles over the image to initialize a color propagation algorithm, or by looking for a suitable reference image from which color information can be transferred. Even with this user supplied data, colorized images may appear unnatural as a result of limited user skill or inaccurate transfer of colors. To address these problems, we propose a colorization system that leverages the rich image content on the internet. As input, the user needs only to provide a semantic text label and segmentation cues for major foreground objects in the scene. With this information, images are downloaded from photo sharing websites and filtered to obtain suitable reference images that are reliable for color transfer to the given grayscale photo. Different image colorizations are generated from the various reference images, and a graphical user interface is provided to easily select the desired result. Our experiments and user study demonstrate the greater effectiveness of this system in comparison to previous techniques.

Training multilayer neural networks using fast global learning algorithm - least-squares and penalized optimization methods

Abstract The major limitations of conventional learning algorithms are attributed to local minima and slow convergence speed. This paper presents a novel heuristics approach for neural networks global learning algorithm. The proposed algorithm is based upon the least-squares (LS) method to maintain the fast convergence speed and a Penalty (PEN) approach to solve the problem of local minima. The penalty term superimposes into the error surface, which likely to provide a way of escape from the local minima when the convergence stalls. The choice and adjustment for the penalty factor are also derived to demonstrate the effect of the penalty term and to ensure the convergence of the algorithm. The developed learning algorithm is applied to several problems of classification application. In all the tested problems, the proposed algorithm outperforms other conventional algorithms in terms of convergence speed and the ability of escaping from the local minima.

Expression recognition using fuzzy spatio-temporal modeling

In human-computer interaction, there is a need for computer to recognize human facial expression accurately. This paper proposes a novel and effective approach for facial expression recognition that analyzes a sequence of images (displaying one expression) instead of just one image (which captures the snapshot of an emotion). Fourier transform is employed to extract features to represent an expression. The representation is further processed using the fuzzy C means computation to generate a spatio-temporal model for each expression type. Unknown input expressions are matched to the models using the Hausdorff distance to compute dissimilarity values for classification. The proposed technique has been tested with the CMU expression database, generating superior results as compared to other approaches.

An improved algorithm for learning long-term dependency problems in adaptive processing of data structures

Many researchers have explored the use of neural-network representations for the adaptive processing of data structures. One of the most popular learning formulations of data structure processing is backpropagation through structure (BPTS). The BPTS algorithm has been successful applied to a number of learning tasks that involve structural patterns such as logo and natural scene classification. The main limitations of the BPTS algorithm are attributed to slow convergence speed and the long-term dependency problem for the adaptive processing of data structures. In this paper, an improved algorithm is proposed to solve these problems. The idea of this algorithm is to optimize the free learning parameters of the neural network in the node representation by using least-squares-based optimization methods in a layer-by-layer fashion. Not only can fast convergence speed be achieved, but the long-term dependency problem can also be overcome since the vanishing of gradient information is avoided when our approach is applied to very deep tree structures.

A novel framework for making dominant point detection methods non-parametric

Most dominant point detection methods require heuristically chosen control parameters. One of the commonly used control parameter is maximum deviation. This paper uses a theoretical bound of the maximum deviation of pixels obtained by digitization of a line segment for constructing a general framework to make most dominant point detection methods non-parametric. The derived analytical bound of the maximum deviation can be used as a natural bench mark for the line fitting algorithms and thus dominant point detection methods can be made parameter-independent and non-heuristic. Most methods can easily incorporate the bound. This is demonstrated using three categorically different dominant point detection methods. Such non-parametric approach retains the characteristics of the digital curve while providing good fitting performance and compression ratio for all the three methods using a variety of digital, non-digital, and noisy curves.

Genetic evolution processing of data structures for image classification

This paper describes a method of structural pattern recognition based on a genetic evolution processing of data structures with neural networks representation. Conventionally, one of the most popular learning formulations of data structure processing is backpropagation through structures (BPTS) [C. Goller et al., (1996)]. The BPTS algorithm has been successfully applied to a number of learning tasks that involved structural patterns such as image, shape, and texture classifications. However, this BPTS typed algorithm suffers from the long-term dependency problem in learning very deep tree structures. In this paper, we propose the genetic evolution for this data structures processing. The idea of this algorithm is to tune the learning parameters by the genetic evolution with specified chromosome structures. Also, the fitness evaluation as well as the adaptive crossover and mutation for this structural genetic processing are investigated in this paper. An application to flowers image classification by a structural representation is provided for the validation of our method. The obtained results significantly support the capabilities of our proposed approach to classify and recognize flowers in terms of generalization and noise robustness.

HebbR2-Taffic: A novel application of neuro-fuzzy network for visual based traffic monitoring system

This paper presents a robust methodology that automatically counts moving vehicles along an expressway. The domain of interest for this paper is using both neuro-fuzzy network and simple image processing techniques to implement traffic flow monitoring and analysis. As this system is dedicated for outdoor applications, efficient and robust processing methods are introduced to handle both day and night analysis. In our study, a neuro-fuzzy network based on the Hebbian-Mamdani rule reduction architecture is used to classify and count the number of vehicles that passed through a three- or four-lanes expressway. As the quality of the video captured is corrupted under noisy outdoor environment, a series of preprocessing is required before the features are fed into the network. A vector of nine feature values is extracted to represent whether a vehicle is passing through a lane and this vector serves as input patterns would be used to train the neuro-fuzzy network. The vehicle counting and classification would then be performed by the well-trained network. The novel approach is benchmarked against the MLP and RBF networks. The results of using our proposed neuro-fuzzy network are very encouraging with a high degree of accuracy.

Neural computation approach for developing a 3D shape reconstruction model

The shape from shading problem refers to the well-known fact that most real images usually contain specular components and are affected by unknown reflectivity. In this paper, these limitations are addressed and a new neural-based 3D shape reconstruction model is proposed. The idea behind this approach is to optimize a proper reflectance model by learning the parameters of the proposed neural reflectance model. In order to do this, new neural-based reflectance models are presented. The feedforward neural network (FNN) model is able to generalize the diffuse term, while the RBF model is able to generalize the specular term. A hybrid structure of FNN-based and RBF-based models is also presented because most real surfaces are usually neither Lambertian models nor ideally specular models. Experimental results, including synthetic and real images, are presented to demonstrate the performance of our approach given different specular effects, unknown illuminate conditions, and different noise environments.

A Fast Neural Learning Vision System for Crowd Estimation at Underground Stations Platform

A neural learning-based crowd estimation system for surveillance in complex scenes at the platform of underground stations is presented. Estimation is carried out by extracting a set of significant features from the sequences of images. Feature indices are modeled by the neural networks to estimate the crowd density. The learning phase is based on our proposed hybrid algorithms which are capable of providing the global search characteristic and fast convergence speed. Promising experimental results were obtained in terms of estimation accuracy and real-time response capability to alert the operators automatically.