T.R. Ren

Image contrast enhancement with brightness preservation using an optimal gamma correction and weighted sum approach

The enhancement of image contrast and preservation of image brightness are two important but conflicting objectives in image restoration. Previous attempts based on linear histogram equalization had achieved contrast enhancement, but exact preservation of brightness was not accomplished. A new perspective is taken here to provide balanced performance of contrast enhancement and brightness preservation simultaneously by casting the quest of such solution to an optimization problem. Specifically, the non-linear gamma correction method is adopted to enhance the contrast, while a weighted sum approach is employed for brightness preservation. In addition, the efficient golden search algorithm is exploited to determine the required optimal parameters to produce the enhanced images. Experiments are conducted on natural colour images captured under various indoor, outdoor and illumination conditions. Results have shown that the proposed method outperforms currently available methods in contrast to enhancement and brightness preservation.

A survey on ellipse detection methods

Ellipses and elliptical features are evident in abundance, in a wide variety of digital images. Much of these features carry within itself useful statistical and geometrical information that can be exploited for a broad range of real-world applications. Algorithms developed of late for ellipse detection are application specific and are mainly based on traditional least-square fitting and Hough transform methods. This, in essence, is a step away from building a fully autonomous system with ellipse detection capabilities. This review attempts to redirect the research focus back towards a common goal of generating new ideas through the introduction of a modular framework.

A Restricted Coulomb Energy (RCE) Neural Network System for Hand Image Segmentation

A hands segmentation scheme based on human skin color classification using the Restricted Coulomb Energy (RCE) neural network is proposed in this paper. An improved iteration strategy for the RCE neural network, with a reduction in the number of repetitive calculations, is utilized in our work. The experimental results show that our system is more accurate and less computational expensive than previous schemes.

Controller design of a truck and multiple trailer system

The use of truck-and-trailer systems is an attractive solution to boost the load carrying capacity of land transportation vehicles. However, the steering of such system is problematic due to the lack of sufficient degree-of-freedoms in the available control. The difficulty further increases when obstacles are encountered in the working space. Here, the virtual-robot tracking strategy and force field method are employed such that the truck is steered to follow a desired trajectory without collision with obstacles. The generation of drive speed and turn-rate is tackled as an optimization problem where the particle swarm optimization algorithm is used for its promising performance and implementation simplicity. Simulation studies on a truck-and-trailer system moving on irregular paths in obstacle filled working space are conducted using the developed intelligent controller and satisfactory results have illustrated the feasibility of the suggested approach.

The impact of information volume on SIFT descriptor

This paper provides a performance evaluation on the Scale- invariant Feature Transform (SIFT) descriptors that utilise different sizes of image patches to represent the SIFT keypoints in images. Although SIFT has been widely employed in numerous applications such as object recognition and image registration, its performances against different image complexities and transformations are still unclear. Thus, an evaluation is commenced to examine SIFT descriptors performance while its dimension (i.e., information volume) is varied. This paper is started by providing the general concept of SIFT descriptor, then the experimental setup and evaluation metrics are described for detailing the performance evaluation. The experimental results are shown by two evaluation metrics that are repeatability and recall-precision. Lastly, discussions and conclusions are included to emphasise the significances observed in the experimental results and highlight possible directions for future work.

Image mosaic construction using feature matching and weighted fusion

An image mosaic construction method with focuses in feature matching and image fusion is proposed. The merit of this method is that it can effectively reduce the rate of feature mismatches. Corners in two overlapping images are taken as features which are detected and matched in order to derive a transformation matrix to align the images to be combined. In order to guarantee a reduced rate of mismatching, a similarity definitude algorithm is employed. Furthermore, a smooth mosaic is obtained by the use of a weighted fusion process. Experimental results, which are conducted using indoor images, show that our approach can effectively obtain an accurate and seamless fused image.

Automatic rule turning fuzzy controller design for a truck and trailer system

Fuzzy logic theories, in the past decades, have been used as a popular basis for designing motion controllers. In this research, a fuzzy logic controller is designed for a truck and multiple trailer system where the controller generates the drive velocity and turn rate for only the truck. Instead of relying solely on human expert inputs, an automatic fuzzy rule tuning approach is adopted. This problem is tackled by using the particle swarm optimization algorithm which removes the need for human expert knowledge input while searching for a near optimal solution. The truck and trailer system can thus be driven to a desired position with a near-optimal trajectory in a two-dimensional terrain. Simulation results have shown the effectiveness of the derived fuzzy controller.

A comparison study on human action recognition from video streams

A vision-based smart building control system relies upon human action recognition to determine the number of occupants inside the building and their respective motion type or path. Using this information, a control system that can automatically optimise environmental conditions within the building can be designed. Furthermore, information obtained is also used to aid in other domains such as security and surveillance, interactive application with environment, and content-based video analysis. Current work relating to the recognition task is divided into two processes: human action extraction and human action classification. This paper starts by identifying the distinct difference between global and local extraction methods. The extraction methods filter out features, such as silhouette, colour, edge, motion and interest point, from images for analysing observed human actions. In terms of human action classification, two key methods known as the k-nearest neighbour approach and hidden Markov model are presented and discussed. Lastly, the paper provides a brief summary highlighting gaps and possible milestones for future work.

Hand image segmentation under selective illuminations

A hand image segmentation method with focuses in constructing Gaussians mixture models (GMMs) for hand skin color is proposed. The Expectation Maximization (EM) algorithm is used for estimating parameters for GMM, and the Receiver Operating Characteristic (ROC) curve is employed for determining a threshold. Experimental results show that two thresholds, which are an overall probability threshold and a hue value threshold, are more capable of achieving ideal segmenting results than a single threshold.